blob: 2ccc35c4dc24d95e2a5c1bf5528528fa6f22827d [file] [log] [blame]
Linus Torvalds1da177e2005-04-16 15:20:36 -07001/*
2 * linux/fs/namei.c
3 *
4 * Copyright (C) 1991, 1992 Linus Torvalds
5 */
6
7/*
8 * Some corrections by tytso.
9 */
10
11/* [Feb 1997 T. Schoebel-Theuer] Complete rewrite of the pathname
12 * lookup logic.
13 */
14/* [Feb-Apr 2000, AV] Rewrite to the new namespace architecture.
15 */
16
17#include <linux/init.h>
Paul Gortmaker630d9c42011-11-16 23:57:37 -050018#include <linux/export.h>
David S. Miller44696902012-05-23 20:12:50 -070019#include <linux/kernel.h>
Linus Torvalds1da177e2005-04-16 15:20:36 -070020#include <linux/slab.h>
21#include <linux/fs.h>
22#include <linux/namei.h>
Linus Torvalds1da177e2005-04-16 15:20:36 -070023#include <linux/pagemap.h>
Robert Love0eeca282005-07-12 17:06:03 -040024#include <linux/fsnotify.h>
Linus Torvalds1da177e2005-04-16 15:20:36 -070025#include <linux/personality.h>
26#include <linux/security.h>
Mimi Zohar6146f0d2009-02-04 09:06:57 -050027#include <linux/ima.h>
Linus Torvalds1da177e2005-04-16 15:20:36 -070028#include <linux/syscalls.h>
29#include <linux/mount.h>
30#include <linux/audit.h>
Randy Dunlap16f7e0f2006-01-11 12:17:46 -080031#include <linux/capability.h>
Trond Myklebust834f2a42005-10-18 14:20:16 -070032#include <linux/file.h>
Ulrich Drepper5590ff02006-01-18 17:43:53 -080033#include <linux/fcntl.h>
Serge E. Hallyn08ce5f12008-04-29 01:00:10 -070034#include <linux/device_cgroup.h>
Al Viro5ad4e532009-03-29 19:50:06 -040035#include <linux/fs_struct.h>
Linus Torvaldse77819e2011-07-22 19:30:19 -070036#include <linux/posix_acl.h>
Linus Torvalds1da177e2005-04-16 15:20:36 -070037#include <asm/uaccess.h>
38
Eric Parise81e3f42009-12-04 15:47:36 -050039#include "internal.h"
Al Viroc7105362011-11-24 18:22:03 -050040#include "mount.h"
Eric Parise81e3f42009-12-04 15:47:36 -050041
Linus Torvalds1da177e2005-04-16 15:20:36 -070042/* [Feb-1997 T. Schoebel-Theuer]
43 * Fundamental changes in the pathname lookup mechanisms (namei)
44 * were necessary because of omirr. The reason is that omirr needs
45 * to know the _real_ pathname, not the user-supplied one, in case
46 * of symlinks (and also when transname replacements occur).
47 *
48 * The new code replaces the old recursive symlink resolution with
49 * an iterative one (in case of non-nested symlink chains). It does
50 * this with calls to <fs>_follow_link().
51 * As a side effect, dir_namei(), _namei() and follow_link() are now
52 * replaced with a single function lookup_dentry() that can handle all
53 * the special cases of the former code.
54 *
55 * With the new dcache, the pathname is stored at each inode, at least as
56 * long as the refcount of the inode is positive. As a side effect, the
57 * size of the dcache depends on the inode cache and thus is dynamic.
58 *
59 * [29-Apr-1998 C. Scott Ananian] Updated above description of symlink
60 * resolution to correspond with current state of the code.
61 *
62 * Note that the symlink resolution is not *completely* iterative.
63 * There is still a significant amount of tail- and mid- recursion in
64 * the algorithm. Also, note that <fs>_readlink() is not used in
65 * lookup_dentry(): lookup_dentry() on the result of <fs>_readlink()
66 * may return different results than <fs>_follow_link(). Many virtual
67 * filesystems (including /proc) exhibit this behavior.
68 */
69
70/* [24-Feb-97 T. Schoebel-Theuer] Side effects caused by new implementation:
71 * New symlink semantics: when open() is called with flags O_CREAT | O_EXCL
72 * and the name already exists in form of a symlink, try to create the new
73 * name indicated by the symlink. The old code always complained that the
74 * name already exists, due to not following the symlink even if its target
75 * is nonexistent. The new semantics affects also mknod() and link() when
Lucas De Marchi25985ed2011-03-30 22:57:33 -030076 * the name is a symlink pointing to a non-existent name.
Linus Torvalds1da177e2005-04-16 15:20:36 -070077 *
78 * I don't know which semantics is the right one, since I have no access
79 * to standards. But I found by trial that HP-UX 9.0 has the full "new"
80 * semantics implemented, while SunOS 4.1.1 and Solaris (SunOS 5.4) have the
81 * "old" one. Personally, I think the new semantics is much more logical.
82 * Note that "ln old new" where "new" is a symlink pointing to a non-existing
83 * file does succeed in both HP-UX and SunOs, but not in Solaris
84 * and in the old Linux semantics.
85 */
86
87/* [16-Dec-97 Kevin Buhr] For security reasons, we change some symlink
88 * semantics. See the comments in "open_namei" and "do_link" below.
89 *
90 * [10-Sep-98 Alan Modra] Another symlink change.
91 */
92
93/* [Feb-Apr 2000 AV] Complete rewrite. Rules for symlinks:
94 * inside the path - always follow.
95 * in the last component in creation/removal/renaming - never follow.
96 * if LOOKUP_FOLLOW passed - follow.
97 * if the pathname has trailing slashes - follow.
98 * otherwise - don't follow.
99 * (applied in that order).
100 *
101 * [Jun 2000 AV] Inconsistent behaviour of open() in case if flags==O_CREAT
102 * restored for 2.4. This is the last surviving part of old 4.2BSD bug.
103 * During the 2.4 we need to fix the userland stuff depending on it -
104 * hopefully we will be able to get rid of that wart in 2.5. So far only
105 * XEmacs seems to be relying on it...
106 */
107/*
108 * [Sep 2001 AV] Single-semaphore locking scheme (kudos to David Holland)
Arjan van de Vena11f3a02006-03-23 03:00:33 -0800109 * implemented. Let's see if raised priority of ->s_vfs_rename_mutex gives
Linus Torvalds1da177e2005-04-16 15:20:36 -0700110 * any extra contention...
111 */
112
113/* In order to reduce some races, while at the same time doing additional
114 * checking and hopefully speeding things up, we copy filenames to the
115 * kernel data space before using them..
116 *
117 * POSIX.1 2.4: an empty pathname is invalid (ENOENT).
118 * PATH_MAX includes the nul terminator --RR.
119 */
Andy Whitcroft1fa1e7f2011-11-02 09:44:39 +0100120static char *getname_flags(const char __user *filename, int flags, int *empty)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700121{
Linus Torvalds3f9f0aa2012-04-28 14:38:32 -0700122 char *result = __getname(), *err;
123 int len;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700124
Linus Torvalds3f9f0aa2012-04-28 14:38:32 -0700125 if (unlikely(!result))
Eric Paris4043cde2012-01-03 14:23:08 -0500126 return ERR_PTR(-ENOMEM);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700127
Linus Torvalds3f9f0aa2012-04-28 14:38:32 -0700128 len = strncpy_from_user(result, filename, PATH_MAX);
129 err = ERR_PTR(len);
130 if (unlikely(len < 0))
131 goto error;
132
133 /* The empty path is special. */
134 if (unlikely(!len)) {
135 if (empty)
Eric Paris4043cde2012-01-03 14:23:08 -0500136 *empty = 1;
Linus Torvalds3f9f0aa2012-04-28 14:38:32 -0700137 err = ERR_PTR(-ENOENT);
138 if (!(flags & LOOKUP_EMPTY))
139 goto error;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700140 }
Linus Torvalds3f9f0aa2012-04-28 14:38:32 -0700141
142 err = ERR_PTR(-ENAMETOOLONG);
143 if (likely(len < PATH_MAX)) {
144 audit_getname(result);
145 return result;
146 }
147
148error:
149 __putname(result);
150 return err;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700151}
152
Al Virof52e0c12011-03-14 18:56:51 -0400153char *getname(const char __user * filename)
154{
Linus Torvaldsf7493e52012-03-22 16:10:40 -0700155 return getname_flags(filename, 0, NULL);
Al Virof52e0c12011-03-14 18:56:51 -0400156}
157
Linus Torvalds1da177e2005-04-16 15:20:36 -0700158#ifdef CONFIG_AUDITSYSCALL
159void putname(const char *name)
160{
Al Viro5ac3a9c2006-07-16 06:38:45 -0400161 if (unlikely(!audit_dummy_context()))
Linus Torvalds1da177e2005-04-16 15:20:36 -0700162 audit_putname(name);
163 else
164 __putname(name);
165}
166EXPORT_SYMBOL(putname);
167#endif
168
Linus Torvaldse77819e2011-07-22 19:30:19 -0700169static int check_acl(struct inode *inode, int mask)
170{
Linus Torvalds84635d62011-07-25 22:47:03 -0700171#ifdef CONFIG_FS_POSIX_ACL
Linus Torvaldse77819e2011-07-22 19:30:19 -0700172 struct posix_acl *acl;
173
Linus Torvaldse77819e2011-07-22 19:30:19 -0700174 if (mask & MAY_NOT_BLOCK) {
Al Viro35678662011-08-02 21:32:13 -0400175 acl = get_cached_acl_rcu(inode, ACL_TYPE_ACCESS);
176 if (!acl)
Linus Torvaldse77819e2011-07-22 19:30:19 -0700177 return -EAGAIN;
Al Viro35678662011-08-02 21:32:13 -0400178 /* no ->get_acl() calls in RCU mode... */
179 if (acl == ACL_NOT_CACHED)
180 return -ECHILD;
Ari Savolainen206b1d02011-08-06 19:43:07 +0300181 return posix_acl_permission(inode, acl, mask & ~MAY_NOT_BLOCK);
Linus Torvaldse77819e2011-07-22 19:30:19 -0700182 }
183
184 acl = get_cached_acl(inode, ACL_TYPE_ACCESS);
185
186 /*
Christoph Hellwig4e34e712011-07-23 17:37:31 +0200187 * A filesystem can force a ACL callback by just never filling the
188 * ACL cache. But normally you'd fill the cache either at inode
189 * instantiation time, or on the first ->get_acl call.
Linus Torvaldse77819e2011-07-22 19:30:19 -0700190 *
Christoph Hellwig4e34e712011-07-23 17:37:31 +0200191 * If the filesystem doesn't have a get_acl() function at all, we'll
192 * just create the negative cache entry.
Linus Torvaldse77819e2011-07-22 19:30:19 -0700193 */
194 if (acl == ACL_NOT_CACHED) {
Christoph Hellwig4e34e712011-07-23 17:37:31 +0200195 if (inode->i_op->get_acl) {
196 acl = inode->i_op->get_acl(inode, ACL_TYPE_ACCESS);
197 if (IS_ERR(acl))
198 return PTR_ERR(acl);
199 } else {
200 set_cached_acl(inode, ACL_TYPE_ACCESS, NULL);
201 return -EAGAIN;
202 }
Linus Torvaldse77819e2011-07-22 19:30:19 -0700203 }
204
205 if (acl) {
206 int error = posix_acl_permission(inode, acl, mask);
207 posix_acl_release(acl);
208 return error;
209 }
Linus Torvalds84635d62011-07-25 22:47:03 -0700210#endif
Linus Torvaldse77819e2011-07-22 19:30:19 -0700211
212 return -EAGAIN;
213}
214
Linus Torvalds5909ccaa2009-08-28 11:51:25 -0700215/*
Andreas Gruenbacher948409c2011-10-23 23:13:33 +0530216 * This does the basic permission checking
Linus Torvalds5909ccaa2009-08-28 11:51:25 -0700217 */
Al Viro7e401452011-06-20 19:12:17 -0400218static int acl_permission_check(struct inode *inode, int mask)
Linus Torvalds5909ccaa2009-08-28 11:51:25 -0700219{
Linus Torvalds26cf46b2011-05-13 11:51:01 -0700220 unsigned int mode = inode->i_mode;
Linus Torvalds5909ccaa2009-08-28 11:51:25 -0700221
Eric W. Biederman8e96e3b2012-03-03 21:17:15 -0800222 if (likely(uid_eq(current_fsuid(), inode->i_uid)))
Linus Torvalds5909ccaa2009-08-28 11:51:25 -0700223 mode >>= 6;
224 else {
Linus Torvaldse77819e2011-07-22 19:30:19 -0700225 if (IS_POSIXACL(inode) && (mode & S_IRWXG)) {
Al Viro7e401452011-06-20 19:12:17 -0400226 int error = check_acl(inode, mask);
Nick Pigginb74c79e2011-01-07 17:49:58 +1100227 if (error != -EAGAIN)
228 return error;
Linus Torvalds5909ccaa2009-08-28 11:51:25 -0700229 }
230
231 if (in_group_p(inode->i_gid))
232 mode >>= 3;
233 }
234
235 /*
236 * If the DACs are ok we don't need any capability check.
237 */
Al Viro9c2c7032011-06-20 19:06:22 -0400238 if ((mask & ~mode & (MAY_READ | MAY_WRITE | MAY_EXEC)) == 0)
Linus Torvalds5909ccaa2009-08-28 11:51:25 -0700239 return 0;
240 return -EACCES;
241}
Linus Torvalds1da177e2005-04-16 15:20:36 -0700242
243/**
Nick Pigginb74c79e2011-01-07 17:49:58 +1100244 * generic_permission - check for access rights on a Posix-like filesystem
Linus Torvalds1da177e2005-04-16 15:20:36 -0700245 * @inode: inode to check access rights for
Andreas Gruenbacher8fd90c82011-10-23 23:13:30 +0530246 * @mask: right to check for (%MAY_READ, %MAY_WRITE, %MAY_EXEC, ...)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700247 *
248 * Used to check for read/write/execute permissions on a file.
249 * We use "fsuid" for this, letting us set arbitrary permissions
250 * for filesystem access without changing the "normal" uids which
Nick Pigginb74c79e2011-01-07 17:49:58 +1100251 * are used for other things.
252 *
253 * generic_permission is rcu-walk aware. It returns -ECHILD in case an rcu-walk
254 * request cannot be satisfied (eg. requires blocking or too much complexity).
255 * It would then be called again in ref-walk mode.
Linus Torvalds1da177e2005-04-16 15:20:36 -0700256 */
Al Viro2830ba72011-06-20 19:16:29 -0400257int generic_permission(struct inode *inode, int mask)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700258{
Linus Torvalds5909ccaa2009-08-28 11:51:25 -0700259 int ret;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700260
261 /*
Andreas Gruenbacher948409c2011-10-23 23:13:33 +0530262 * Do the basic permission checks.
Linus Torvalds1da177e2005-04-16 15:20:36 -0700263 */
Al Viro7e401452011-06-20 19:12:17 -0400264 ret = acl_permission_check(inode, mask);
Linus Torvalds5909ccaa2009-08-28 11:51:25 -0700265 if (ret != -EACCES)
266 return ret;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700267
Al Virod594e7e2011-06-20 19:55:42 -0400268 if (S_ISDIR(inode->i_mode)) {
269 /* DACs are overridable for directories */
Eric W. Biederman1a48e2a2011-11-14 16:24:06 -0800270 if (inode_capable(inode, CAP_DAC_OVERRIDE))
Al Virod594e7e2011-06-20 19:55:42 -0400271 return 0;
272 if (!(mask & MAY_WRITE))
Eric W. Biederman1a48e2a2011-11-14 16:24:06 -0800273 if (inode_capable(inode, CAP_DAC_READ_SEARCH))
Al Virod594e7e2011-06-20 19:55:42 -0400274 return 0;
275 return -EACCES;
276 }
Linus Torvalds1da177e2005-04-16 15:20:36 -0700277 /*
278 * Read/write DACs are always overridable.
Al Virod594e7e2011-06-20 19:55:42 -0400279 * Executable DACs are overridable when there is
280 * at least one exec bit set.
Linus Torvalds1da177e2005-04-16 15:20:36 -0700281 */
Al Virod594e7e2011-06-20 19:55:42 -0400282 if (!(mask & MAY_EXEC) || (inode->i_mode & S_IXUGO))
Eric W. Biederman1a48e2a2011-11-14 16:24:06 -0800283 if (inode_capable(inode, CAP_DAC_OVERRIDE))
Linus Torvalds1da177e2005-04-16 15:20:36 -0700284 return 0;
285
286 /*
287 * Searching includes executable on directories, else just read.
288 */
Serge E. Hallyn7ea66002009-12-29 14:50:19 -0600289 mask &= MAY_READ | MAY_WRITE | MAY_EXEC;
Al Virod594e7e2011-06-20 19:55:42 -0400290 if (mask == MAY_READ)
Eric W. Biederman1a48e2a2011-11-14 16:24:06 -0800291 if (inode_capable(inode, CAP_DAC_READ_SEARCH))
Linus Torvalds1da177e2005-04-16 15:20:36 -0700292 return 0;
293
294 return -EACCES;
295}
296
Linus Torvalds3ddcd052011-08-06 22:45:50 -0700297/*
298 * We _really_ want to just do "generic_permission()" without
299 * even looking at the inode->i_op values. So we keep a cache
300 * flag in inode->i_opflags, that says "this has not special
301 * permission function, use the fast case".
302 */
303static inline int do_inode_permission(struct inode *inode, int mask)
304{
305 if (unlikely(!(inode->i_opflags & IOP_FASTPERM))) {
306 if (likely(inode->i_op->permission))
307 return inode->i_op->permission(inode, mask);
308
309 /* This gets set once for the inode lifetime */
310 spin_lock(&inode->i_lock);
311 inode->i_opflags |= IOP_FASTPERM;
312 spin_unlock(&inode->i_lock);
313 }
314 return generic_permission(inode, mask);
315}
316
Christoph Hellwigcb23beb2008-10-24 09:59:29 +0200317/**
David Howells0bdaea92012-06-25 12:55:46 +0100318 * __inode_permission - Check for access rights to a given inode
319 * @inode: Inode to check permission on
320 * @mask: Right to check for (%MAY_READ, %MAY_WRITE, %MAY_EXEC)
Christoph Hellwigcb23beb2008-10-24 09:59:29 +0200321 *
David Howells0bdaea92012-06-25 12:55:46 +0100322 * Check for read/write/execute permissions on an inode.
Andreas Gruenbacher948409c2011-10-23 23:13:33 +0530323 *
324 * When checking for MAY_APPEND, MAY_WRITE must also be set in @mask.
David Howells0bdaea92012-06-25 12:55:46 +0100325 *
326 * This does not check for a read-only file system. You probably want
327 * inode_permission().
Christoph Hellwigcb23beb2008-10-24 09:59:29 +0200328 */
David Howells0bdaea92012-06-25 12:55:46 +0100329int __inode_permission(struct inode *inode, int mask)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700330{
Al Viroe6305c42008-07-15 21:03:57 -0400331 int retval;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700332
Linus Torvalds3ddcd052011-08-06 22:45:50 -0700333 if (unlikely(mask & MAY_WRITE)) {
Linus Torvalds1da177e2005-04-16 15:20:36 -0700334 /*
335 * Nobody gets write access to an immutable file.
336 */
337 if (IS_IMMUTABLE(inode))
338 return -EACCES;
339 }
340
Linus Torvalds3ddcd052011-08-06 22:45:50 -0700341 retval = do_inode_permission(inode, mask);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700342 if (retval)
343 return retval;
344
Serge E. Hallyn08ce5f12008-04-29 01:00:10 -0700345 retval = devcgroup_inode_permission(inode, mask);
346 if (retval)
347 return retval;
348
Eric Parisd09ca732010-07-23 11:43:57 -0400349 return security_inode_permission(inode, mask);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700350}
351
Al Virof4d6ff82011-06-19 13:14:21 -0400352/**
David Howells0bdaea92012-06-25 12:55:46 +0100353 * sb_permission - Check superblock-level permissions
354 * @sb: Superblock of inode to check permission on
355 * @mask: Right to check for (%MAY_READ, %MAY_WRITE, %MAY_EXEC)
356 *
357 * Separate out file-system wide checks from inode-specific permission checks.
358 */
359static int sb_permission(struct super_block *sb, struct inode *inode, int mask)
360{
361 if (unlikely(mask & MAY_WRITE)) {
362 umode_t mode = inode->i_mode;
363
364 /* Nobody gets write access to a read-only fs. */
365 if ((sb->s_flags & MS_RDONLY) &&
366 (S_ISREG(mode) || S_ISDIR(mode) || S_ISLNK(mode)))
367 return -EROFS;
368 }
369 return 0;
370}
371
372/**
373 * inode_permission - Check for access rights to a given inode
374 * @inode: Inode to check permission on
375 * @mask: Right to check for (%MAY_READ, %MAY_WRITE, %MAY_EXEC)
376 *
377 * Check for read/write/execute permissions on an inode. We use fs[ug]id for
378 * this, letting us set arbitrary permissions for filesystem access without
379 * changing the "normal" UIDs which are used for other things.
380 *
381 * When checking for MAY_APPEND, MAY_WRITE must also be set in @mask.
382 */
383int inode_permission(struct inode *inode, int mask)
384{
385 int retval;
386
387 retval = sb_permission(inode->i_sb, inode, mask);
388 if (retval)
389 return retval;
390 return __inode_permission(inode, mask);
391}
392
393/**
Jan Blunck5dd784d02008-02-14 19:34:38 -0800394 * path_get - get a reference to a path
395 * @path: path to get the reference to
396 *
397 * Given a path increment the reference count to the dentry and the vfsmount.
398 */
399void path_get(struct path *path)
400{
401 mntget(path->mnt);
402 dget(path->dentry);
403}
404EXPORT_SYMBOL(path_get);
405
406/**
Jan Blunck1d957f92008-02-14 19:34:35 -0800407 * path_put - put a reference to a path
408 * @path: path to put the reference to
409 *
410 * Given a path decrement the reference count to the dentry and the vfsmount.
411 */
412void path_put(struct path *path)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700413{
Jan Blunck1d957f92008-02-14 19:34:35 -0800414 dput(path->dentry);
415 mntput(path->mnt);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700416}
Jan Blunck1d957f92008-02-14 19:34:35 -0800417EXPORT_SYMBOL(path_put);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700418
Al Viro19660af2011-03-25 10:32:48 -0400419/*
Nick Piggin31e6b012011-01-07 17:49:52 +1100420 * Path walking has 2 modes, rcu-walk and ref-walk (see
Al Viro19660af2011-03-25 10:32:48 -0400421 * Documentation/filesystems/path-lookup.txt). In situations when we can't
422 * continue in RCU mode, we attempt to drop out of rcu-walk mode and grab
423 * normal reference counts on dentries and vfsmounts to transition to rcu-walk
424 * mode. Refcounts are grabbed at the last known good point before rcu-walk
425 * got stuck, so ref-walk may continue from there. If this is not successful
426 * (eg. a seqcount has changed), then failure is returned and it's up to caller
427 * to restart the path walk from the beginning in ref-walk mode.
Nick Piggin31e6b012011-01-07 17:49:52 +1100428 */
Nick Piggin31e6b012011-01-07 17:49:52 +1100429
Al Viro32a79912012-07-18 20:43:19 +0400430static inline void lock_rcu_walk(void)
431{
432 br_read_lock(&vfsmount_lock);
433 rcu_read_lock();
434}
435
436static inline void unlock_rcu_walk(void)
437{
438 rcu_read_unlock();
439 br_read_unlock(&vfsmount_lock);
440}
441
Nick Piggin31e6b012011-01-07 17:49:52 +1100442/**
Al Viro19660af2011-03-25 10:32:48 -0400443 * unlazy_walk - try to switch to ref-walk mode.
444 * @nd: nameidata pathwalk data
445 * @dentry: child of nd->path.dentry or NULL
Randy Dunlap39191622011-01-08 19:36:21 -0800446 * Returns: 0 on success, -ECHILD on failure
Nick Piggin31e6b012011-01-07 17:49:52 +1100447 *
Al Viro19660af2011-03-25 10:32:48 -0400448 * unlazy_walk attempts to legitimize the current nd->path, nd->root and dentry
449 * for ref-walk mode. @dentry must be a path found by a do_lookup call on
450 * @nd or NULL. Must be called from rcu-walk context.
Nick Piggin31e6b012011-01-07 17:49:52 +1100451 */
Al Viro19660af2011-03-25 10:32:48 -0400452static int unlazy_walk(struct nameidata *nd, struct dentry *dentry)
Nick Piggin31e6b012011-01-07 17:49:52 +1100453{
454 struct fs_struct *fs = current->fs;
455 struct dentry *parent = nd->path.dentry;
Al Viro5b6ca022011-03-09 23:04:47 -0500456 int want_root = 0;
Nick Piggin31e6b012011-01-07 17:49:52 +1100457
458 BUG_ON(!(nd->flags & LOOKUP_RCU));
Al Viro5b6ca022011-03-09 23:04:47 -0500459 if (nd->root.mnt && !(nd->flags & LOOKUP_ROOT)) {
460 want_root = 1;
Nick Piggin31e6b012011-01-07 17:49:52 +1100461 spin_lock(&fs->lock);
462 if (nd->root.mnt != fs->root.mnt ||
463 nd->root.dentry != fs->root.dentry)
464 goto err_root;
465 }
466 spin_lock(&parent->d_lock);
Al Viro19660af2011-03-25 10:32:48 -0400467 if (!dentry) {
468 if (!__d_rcu_to_refcount(parent, nd->seq))
469 goto err_parent;
470 BUG_ON(nd->inode != parent->d_inode);
471 } else {
Al Viro94c0d4e2011-07-12 21:40:23 -0400472 if (dentry->d_parent != parent)
473 goto err_parent;
Al Viro19660af2011-03-25 10:32:48 -0400474 spin_lock_nested(&dentry->d_lock, DENTRY_D_LOCK_NESTED);
475 if (!__d_rcu_to_refcount(dentry, nd->seq))
476 goto err_child;
477 /*
478 * If the sequence check on the child dentry passed, then
479 * the child has not been removed from its parent. This
480 * means the parent dentry must be valid and able to take
481 * a reference at this point.
482 */
483 BUG_ON(!IS_ROOT(dentry) && dentry->d_parent != parent);
484 BUG_ON(!parent->d_count);
485 parent->d_count++;
486 spin_unlock(&dentry->d_lock);
487 }
Nick Piggin31e6b012011-01-07 17:49:52 +1100488 spin_unlock(&parent->d_lock);
Al Viro5b6ca022011-03-09 23:04:47 -0500489 if (want_root) {
Nick Piggin31e6b012011-01-07 17:49:52 +1100490 path_get(&nd->root);
491 spin_unlock(&fs->lock);
492 }
493 mntget(nd->path.mnt);
494
Al Viro32a79912012-07-18 20:43:19 +0400495 unlock_rcu_walk();
Nick Piggin31e6b012011-01-07 17:49:52 +1100496 nd->flags &= ~LOOKUP_RCU;
497 return 0;
Al Viro19660af2011-03-25 10:32:48 -0400498
499err_child:
Nick Piggin31e6b012011-01-07 17:49:52 +1100500 spin_unlock(&dentry->d_lock);
Al Viro19660af2011-03-25 10:32:48 -0400501err_parent:
Nick Piggin31e6b012011-01-07 17:49:52 +1100502 spin_unlock(&parent->d_lock);
503err_root:
Al Viro5b6ca022011-03-09 23:04:47 -0500504 if (want_root)
Nick Piggin31e6b012011-01-07 17:49:52 +1100505 spin_unlock(&fs->lock);
506 return -ECHILD;
507}
508
Al Viro4ce16ef32012-06-10 16:10:59 -0400509static inline int d_revalidate(struct dentry *dentry, unsigned int flags)
Nick Piggin34286d62011-01-07 17:49:57 +1100510{
Al Viro4ce16ef32012-06-10 16:10:59 -0400511 return dentry->d_op->d_revalidate(dentry, flags);
Nick Piggin34286d62011-01-07 17:49:57 +1100512}
513
Al Viro9f1fafe2011-03-25 11:00:12 -0400514/**
515 * complete_walk - successful completion of path walk
516 * @nd: pointer nameidata
Jeff Layton39159de2009-12-07 12:01:50 -0500517 *
Al Viro9f1fafe2011-03-25 11:00:12 -0400518 * If we had been in RCU mode, drop out of it and legitimize nd->path.
519 * Revalidate the final result, unless we'd already done that during
520 * the path walk or the filesystem doesn't ask for it. Return 0 on
521 * success, -error on failure. In case of failure caller does not
522 * need to drop nd->path.
Jeff Layton39159de2009-12-07 12:01:50 -0500523 */
Al Viro9f1fafe2011-03-25 11:00:12 -0400524static int complete_walk(struct nameidata *nd)
Jeff Layton39159de2009-12-07 12:01:50 -0500525{
Al Viro16c2cd72011-02-22 15:50:10 -0500526 struct dentry *dentry = nd->path.dentry;
Jeff Layton39159de2009-12-07 12:01:50 -0500527 int status;
Jeff Layton39159de2009-12-07 12:01:50 -0500528
Al Viro9f1fafe2011-03-25 11:00:12 -0400529 if (nd->flags & LOOKUP_RCU) {
530 nd->flags &= ~LOOKUP_RCU;
531 if (!(nd->flags & LOOKUP_ROOT))
532 nd->root.mnt = NULL;
533 spin_lock(&dentry->d_lock);
534 if (unlikely(!__d_rcu_to_refcount(dentry, nd->seq))) {
535 spin_unlock(&dentry->d_lock);
Al Viro32a79912012-07-18 20:43:19 +0400536 unlock_rcu_walk();
Al Viro9f1fafe2011-03-25 11:00:12 -0400537 return -ECHILD;
538 }
539 BUG_ON(nd->inode != dentry->d_inode);
540 spin_unlock(&dentry->d_lock);
541 mntget(nd->path.mnt);
Al Viro32a79912012-07-18 20:43:19 +0400542 unlock_rcu_walk();
Al Viro9f1fafe2011-03-25 11:00:12 -0400543 }
544
Al Viro16c2cd72011-02-22 15:50:10 -0500545 if (likely(!(nd->flags & LOOKUP_JUMPED)))
Jeff Layton39159de2009-12-07 12:01:50 -0500546 return 0;
547
Al Viro16c2cd72011-02-22 15:50:10 -0500548 if (likely(!(dentry->d_flags & DCACHE_OP_REVALIDATE)))
549 return 0;
550
551 if (likely(!(dentry->d_sb->s_type->fs_flags & FS_REVAL_DOT)))
552 return 0;
553
554 /* Note: we do not d_invalidate() */
Al Viro4ce16ef32012-06-10 16:10:59 -0400555 status = d_revalidate(dentry, nd->flags);
Jeff Layton39159de2009-12-07 12:01:50 -0500556 if (status > 0)
557 return 0;
558
Al Viro16c2cd72011-02-22 15:50:10 -0500559 if (!status)
Jeff Layton39159de2009-12-07 12:01:50 -0500560 status = -ESTALE;
Al Viro16c2cd72011-02-22 15:50:10 -0500561
Al Viro9f1fafe2011-03-25 11:00:12 -0400562 path_put(&nd->path);
Jeff Layton39159de2009-12-07 12:01:50 -0500563 return status;
564}
565
Al Viro2a737872009-04-07 11:49:53 -0400566static __always_inline void set_root(struct nameidata *nd)
567{
Miklos Szeredif7ad3c62010-08-10 11:41:36 +0200568 if (!nd->root.mnt)
569 get_fs_root(current->fs, &nd->root);
Al Viro2a737872009-04-07 11:49:53 -0400570}
571
Al Viro6de88d72009-08-09 01:41:57 +0400572static int link_path_walk(const char *, struct nameidata *);
573
Nick Piggin31e6b012011-01-07 17:49:52 +1100574static __always_inline void set_root_rcu(struct nameidata *nd)
575{
576 if (!nd->root.mnt) {
577 struct fs_struct *fs = current->fs;
Nick Pigginc28cc362011-01-07 17:49:53 +1100578 unsigned seq;
579
580 do {
581 seq = read_seqcount_begin(&fs->seq);
582 nd->root = fs->root;
Tim Chenc1530012011-04-15 11:39:29 -0700583 nd->seq = __read_seqcount_begin(&nd->root.dentry->d_seq);
Nick Pigginc28cc362011-01-07 17:49:53 +1100584 } while (read_seqcount_retry(&fs->seq, seq));
Nick Piggin31e6b012011-01-07 17:49:52 +1100585 }
586}
587
Arjan van de Venf1662352006-01-14 13:21:31 -0800588static __always_inline int __vfs_follow_link(struct nameidata *nd, const char *link)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700589{
Nick Piggin31e6b012011-01-07 17:49:52 +1100590 int ret;
591
Linus Torvalds1da177e2005-04-16 15:20:36 -0700592 if (IS_ERR(link))
593 goto fail;
594
595 if (*link == '/') {
Al Viro2a737872009-04-07 11:49:53 -0400596 set_root(nd);
Jan Blunck1d957f92008-02-14 19:34:35 -0800597 path_put(&nd->path);
Al Viro2a737872009-04-07 11:49:53 -0400598 nd->path = nd->root;
599 path_get(&nd->root);
Al Viro16c2cd72011-02-22 15:50:10 -0500600 nd->flags |= LOOKUP_JUMPED;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700601 }
Nick Piggin31e6b012011-01-07 17:49:52 +1100602 nd->inode = nd->path.dentry->d_inode;
Christoph Hellwigb4091d52008-11-05 15:07:21 +0100603
Nick Piggin31e6b012011-01-07 17:49:52 +1100604 ret = link_path_walk(link, nd);
605 return ret;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700606fail:
Jan Blunck1d957f92008-02-14 19:34:35 -0800607 path_put(&nd->path);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700608 return PTR_ERR(link);
609}
610
Jan Blunck1d957f92008-02-14 19:34:35 -0800611static void path_put_conditional(struct path *path, struct nameidata *nd)
Miklos Szeredi09dd17d2005-09-06 15:18:21 -0700612{
613 dput(path->dentry);
Jan Blunck4ac91372008-02-14 19:34:32 -0800614 if (path->mnt != nd->path.mnt)
Miklos Szeredi09dd17d2005-09-06 15:18:21 -0700615 mntput(path->mnt);
616}
617
Nick Piggin7b9337a2011-01-14 08:42:43 +0000618static inline void path_to_nameidata(const struct path *path,
619 struct nameidata *nd)
Miklos Szeredi09dd17d2005-09-06 15:18:21 -0700620{
Nick Piggin31e6b012011-01-07 17:49:52 +1100621 if (!(nd->flags & LOOKUP_RCU)) {
622 dput(nd->path.dentry);
623 if (nd->path.mnt != path->mnt)
624 mntput(nd->path.mnt);
Huang Shijie9a229682010-04-02 17:37:13 +0800625 }
Nick Piggin31e6b012011-01-07 17:49:52 +1100626 nd->path.mnt = path->mnt;
Jan Blunck4ac91372008-02-14 19:34:32 -0800627 nd->path.dentry = path->dentry;
Miklos Szeredi09dd17d2005-09-06 15:18:21 -0700628}
629
Christoph Hellwigb5fb63c12012-06-18 10:47:04 -0400630/*
631 * Helper to directly jump to a known parsed path from ->follow_link,
632 * caller must have taken a reference to path beforehand.
633 */
634void nd_jump_link(struct nameidata *nd, struct path *path)
635{
636 path_put(&nd->path);
637
638 nd->path = *path;
639 nd->inode = nd->path.dentry->d_inode;
640 nd->flags |= LOOKUP_JUMPED;
641
642 BUG_ON(nd->inode->i_op->follow_link);
643}
644
Al Viro574197e2011-03-14 22:20:34 -0400645static inline void put_link(struct nameidata *nd, struct path *link, void *cookie)
646{
647 struct inode *inode = link->dentry->d_inode;
Al Viro6d7b5aa2012-06-10 04:15:17 -0400648 if (inode->i_op->put_link)
Al Viro574197e2011-03-14 22:20:34 -0400649 inode->i_op->put_link(link->dentry, nd, cookie);
650 path_put(link);
651}
652
Al Virodef4af32009-12-26 08:37:05 -0500653static __always_inline int
Al Viro574197e2011-03-14 22:20:34 -0400654follow_link(struct path *link, struct nameidata *nd, void **p)
Ian Kent051d3812006-03-27 01:14:53 -0800655{
Nick Piggin7b9337a2011-01-14 08:42:43 +0000656 struct dentry *dentry = link->dentry;
Al Viro6d7b5aa2012-06-10 04:15:17 -0400657 int error;
658 char *s;
Ian Kent051d3812006-03-27 01:14:53 -0800659
Al Viro844a3912011-02-15 00:38:26 -0500660 BUG_ON(nd->flags & LOOKUP_RCU);
661
Al Viro0e794582011-03-16 02:45:02 -0400662 if (link->mnt == nd->path.mnt)
663 mntget(link->mnt);
664
Al Viro6d7b5aa2012-06-10 04:15:17 -0400665 error = -ELOOP;
666 if (unlikely(current->total_link_count >= 40))
667 goto out_put_nd_path;
668
Al Viro574197e2011-03-14 22:20:34 -0400669 cond_resched();
670 current->total_link_count++;
671
Al Viro68ac1232012-03-15 08:21:57 -0400672 touch_atime(link);
Ian Kent051d3812006-03-27 01:14:53 -0800673 nd_set_link(nd, NULL);
674
Al Viro36f3b4f2011-02-22 21:24:38 -0500675 error = security_inode_follow_link(link->dentry, nd);
Al Viro6d7b5aa2012-06-10 04:15:17 -0400676 if (error)
677 goto out_put_nd_path;
Al Viro36f3b4f2011-02-22 21:24:38 -0500678
Al Viro86acdca12009-12-22 23:45:11 -0500679 nd->last_type = LAST_BIND;
Al Virodef4af32009-12-26 08:37:05 -0500680 *p = dentry->d_inode->i_op->follow_link(dentry, nd);
681 error = PTR_ERR(*p);
Al Viro6d7b5aa2012-06-10 04:15:17 -0400682 if (IS_ERR(*p))
Christoph Hellwig408ef012012-06-18 10:47:03 -0400683 goto out_put_nd_path;
Al Viro6d7b5aa2012-06-10 04:15:17 -0400684
685 error = 0;
686 s = nd_get_link(nd);
687 if (s) {
688 error = __vfs_follow_link(nd, s);
Christoph Hellwigb5fb63c12012-06-18 10:47:04 -0400689 if (unlikely(error))
690 put_link(nd, link, *p);
Ian Kent051d3812006-03-27 01:14:53 -0800691 }
Al Viro6d7b5aa2012-06-10 04:15:17 -0400692
693 return error;
694
695out_put_nd_path:
696 path_put(&nd->path);
Al Viro6d7b5aa2012-06-10 04:15:17 -0400697 path_put(link);
Ian Kent051d3812006-03-27 01:14:53 -0800698 return error;
699}
700
Nick Piggin31e6b012011-01-07 17:49:52 +1100701static int follow_up_rcu(struct path *path)
702{
Al Viro0714a532011-11-24 22:19:58 -0500703 struct mount *mnt = real_mount(path->mnt);
704 struct mount *parent;
Nick Piggin31e6b012011-01-07 17:49:52 +1100705 struct dentry *mountpoint;
706
Al Viro0714a532011-11-24 22:19:58 -0500707 parent = mnt->mnt_parent;
708 if (&parent->mnt == path->mnt)
Nick Piggin31e6b012011-01-07 17:49:52 +1100709 return 0;
Al Viroa73324d2011-11-24 22:25:07 -0500710 mountpoint = mnt->mnt_mountpoint;
Nick Piggin31e6b012011-01-07 17:49:52 +1100711 path->dentry = mountpoint;
Al Viro0714a532011-11-24 22:19:58 -0500712 path->mnt = &parent->mnt;
Nick Piggin31e6b012011-01-07 17:49:52 +1100713 return 1;
714}
715
David Howellsf015f1262012-06-25 12:55:28 +0100716/*
717 * follow_up - Find the mountpoint of path's vfsmount
718 *
719 * Given a path, find the mountpoint of its source file system.
720 * Replace @path with the path of the mountpoint in the parent mount.
721 * Up is towards /.
722 *
723 * Return 1 if we went up a level and 0 if we were already at the
724 * root.
725 */
Al Virobab77eb2009-04-18 03:26:48 -0400726int follow_up(struct path *path)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700727{
Al Viro0714a532011-11-24 22:19:58 -0500728 struct mount *mnt = real_mount(path->mnt);
729 struct mount *parent;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700730 struct dentry *mountpoint;
Nick Piggin99b7db72010-08-18 04:37:39 +1000731
Andi Kleen962830d2012-05-08 13:32:02 +0930732 br_read_lock(&vfsmount_lock);
Al Viro0714a532011-11-24 22:19:58 -0500733 parent = mnt->mnt_parent;
Al Viro3c0a6162012-07-18 17:32:50 +0400734 if (parent == mnt) {
Andi Kleen962830d2012-05-08 13:32:02 +0930735 br_read_unlock(&vfsmount_lock);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700736 return 0;
737 }
Al Viro0714a532011-11-24 22:19:58 -0500738 mntget(&parent->mnt);
Al Viroa73324d2011-11-24 22:25:07 -0500739 mountpoint = dget(mnt->mnt_mountpoint);
Andi Kleen962830d2012-05-08 13:32:02 +0930740 br_read_unlock(&vfsmount_lock);
Al Virobab77eb2009-04-18 03:26:48 -0400741 dput(path->dentry);
742 path->dentry = mountpoint;
743 mntput(path->mnt);
Al Viro0714a532011-11-24 22:19:58 -0500744 path->mnt = &parent->mnt;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700745 return 1;
746}
747
Nick Pigginb5c84bf2011-01-07 17:49:38 +1100748/*
David Howells9875cf82011-01-14 18:45:21 +0000749 * Perform an automount
750 * - return -EISDIR to tell follow_managed() to stop and return the path we
751 * were called with.
Linus Torvalds1da177e2005-04-16 15:20:36 -0700752 */
David Howells9875cf82011-01-14 18:45:21 +0000753static int follow_automount(struct path *path, unsigned flags,
754 bool *need_mntput)
Nick Piggin31e6b012011-01-07 17:49:52 +1100755{
David Howells9875cf82011-01-14 18:45:21 +0000756 struct vfsmount *mnt;
David Howellsea5b7782011-01-14 19:10:03 +0000757 int err;
Nick Piggin31e6b012011-01-07 17:49:52 +1100758
David Howells9875cf82011-01-14 18:45:21 +0000759 if (!path->dentry->d_op || !path->dentry->d_op->d_automount)
760 return -EREMOTE;
Al Viro463ffb22005-06-06 13:36:05 -0700761
Miklos Szeredi0ec26fd2011-09-05 18:06:26 +0200762 /* We don't want to mount if someone's just doing a stat -
763 * unless they're stat'ing a directory and appended a '/' to
764 * the name.
765 *
766 * We do, however, want to mount if someone wants to open or
767 * create a file of any type under the mountpoint, wants to
768 * traverse through the mountpoint or wants to open the
769 * mounted directory. Also, autofs may mark negative dentries
770 * as being automount points. These will need the attentions
771 * of the daemon to instantiate them before they can be used.
David Howells9875cf82011-01-14 18:45:21 +0000772 */
Miklos Szeredi0ec26fd2011-09-05 18:06:26 +0200773 if (!(flags & (LOOKUP_PARENT | LOOKUP_DIRECTORY |
Linus Torvaldsd94c1772011-09-26 17:44:55 -0700774 LOOKUP_OPEN | LOOKUP_CREATE | LOOKUP_AUTOMOUNT)) &&
Miklos Szeredi0ec26fd2011-09-05 18:06:26 +0200775 path->dentry->d_inode)
776 return -EISDIR;
777
David Howells9875cf82011-01-14 18:45:21 +0000778 current->total_link_count++;
779 if (current->total_link_count >= 40)
780 return -ELOOP;
781
782 mnt = path->dentry->d_op->d_automount(path);
783 if (IS_ERR(mnt)) {
784 /*
785 * The filesystem is allowed to return -EISDIR here to indicate
786 * it doesn't want to automount. For instance, autofs would do
787 * this so that its userspace daemon can mount on this dentry.
788 *
789 * However, we can only permit this if it's a terminal point in
790 * the path being looked up; if it wasn't then the remainder of
791 * the path is inaccessible and we should say so.
792 */
Al Viro49084c32011-06-25 21:59:52 -0400793 if (PTR_ERR(mnt) == -EISDIR && (flags & LOOKUP_PARENT))
David Howells9875cf82011-01-14 18:45:21 +0000794 return -EREMOTE;
795 return PTR_ERR(mnt);
796 }
David Howellsea5b7782011-01-14 19:10:03 +0000797
David Howells9875cf82011-01-14 18:45:21 +0000798 if (!mnt) /* mount collision */
799 return 0;
800
Al Viro8aef1882011-06-16 15:10:06 +0100801 if (!*need_mntput) {
802 /* lock_mount() may release path->mnt on error */
803 mntget(path->mnt);
804 *need_mntput = true;
805 }
Al Viro19a167a2011-01-17 01:35:23 -0500806 err = finish_automount(mnt, path);
David Howellsea5b7782011-01-14 19:10:03 +0000807
David Howellsea5b7782011-01-14 19:10:03 +0000808 switch (err) {
809 case -EBUSY:
810 /* Someone else made a mount here whilst we were busy */
Al Viro19a167a2011-01-17 01:35:23 -0500811 return 0;
David Howellsea5b7782011-01-14 19:10:03 +0000812 case 0:
Al Viro8aef1882011-06-16 15:10:06 +0100813 path_put(path);
David Howellsea5b7782011-01-14 19:10:03 +0000814 path->mnt = mnt;
815 path->dentry = dget(mnt->mnt_root);
David Howellsea5b7782011-01-14 19:10:03 +0000816 return 0;
Al Viro19a167a2011-01-17 01:35:23 -0500817 default:
818 return err;
David Howellsea5b7782011-01-14 19:10:03 +0000819 }
Al Viro19a167a2011-01-17 01:35:23 -0500820
David Howells9875cf82011-01-14 18:45:21 +0000821}
822
823/*
824 * Handle a dentry that is managed in some way.
David Howellscc53ce52011-01-14 18:45:26 +0000825 * - Flagged for transit management (autofs)
David Howells9875cf82011-01-14 18:45:21 +0000826 * - Flagged as mountpoint
827 * - Flagged as automount point
828 *
829 * This may only be called in refwalk mode.
830 *
831 * Serialization is taken care of in namespace.c
832 */
833static int follow_managed(struct path *path, unsigned flags)
834{
Al Viro8aef1882011-06-16 15:10:06 +0100835 struct vfsmount *mnt = path->mnt; /* held by caller, must be left alone */
David Howells9875cf82011-01-14 18:45:21 +0000836 unsigned managed;
837 bool need_mntput = false;
Al Viro8aef1882011-06-16 15:10:06 +0100838 int ret = 0;
David Howells9875cf82011-01-14 18:45:21 +0000839
840 /* Given that we're not holding a lock here, we retain the value in a
841 * local variable for each dentry as we look at it so that we don't see
842 * the components of that value change under us */
843 while (managed = ACCESS_ONCE(path->dentry->d_flags),
844 managed &= DCACHE_MANAGED_DENTRY,
845 unlikely(managed != 0)) {
David Howellscc53ce52011-01-14 18:45:26 +0000846 /* Allow the filesystem to manage the transit without i_mutex
847 * being held. */
848 if (managed & DCACHE_MANAGE_TRANSIT) {
849 BUG_ON(!path->dentry->d_op);
850 BUG_ON(!path->dentry->d_op->d_manage);
Al Viro1aed3e42011-03-18 09:09:02 -0400851 ret = path->dentry->d_op->d_manage(path->dentry, false);
David Howellscc53ce52011-01-14 18:45:26 +0000852 if (ret < 0)
Al Viro8aef1882011-06-16 15:10:06 +0100853 break;
David Howellscc53ce52011-01-14 18:45:26 +0000854 }
855
David Howells9875cf82011-01-14 18:45:21 +0000856 /* Transit to a mounted filesystem. */
857 if (managed & DCACHE_MOUNTED) {
858 struct vfsmount *mounted = lookup_mnt(path);
859 if (mounted) {
860 dput(path->dentry);
861 if (need_mntput)
862 mntput(path->mnt);
863 path->mnt = mounted;
864 path->dentry = dget(mounted->mnt_root);
865 need_mntput = true;
866 continue;
867 }
868
869 /* Something is mounted on this dentry in another
870 * namespace and/or whatever was mounted there in this
871 * namespace got unmounted before we managed to get the
872 * vfsmount_lock */
873 }
874
875 /* Handle an automount point */
876 if (managed & DCACHE_NEED_AUTOMOUNT) {
877 ret = follow_automount(path, flags, &need_mntput);
878 if (ret < 0)
Al Viro8aef1882011-06-16 15:10:06 +0100879 break;
David Howells9875cf82011-01-14 18:45:21 +0000880 continue;
881 }
882
883 /* We didn't change the current path point */
884 break;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700885 }
Al Viro8aef1882011-06-16 15:10:06 +0100886
887 if (need_mntput && path->mnt == mnt)
888 mntput(path->mnt);
889 if (ret == -EISDIR)
890 ret = 0;
Al Viroa3fbbde2011-11-07 21:21:26 +0000891 return ret < 0 ? ret : need_mntput;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700892}
893
David Howellscc53ce52011-01-14 18:45:26 +0000894int follow_down_one(struct path *path)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700895{
896 struct vfsmount *mounted;
897
Al Viro1c755af2009-04-18 14:06:57 -0400898 mounted = lookup_mnt(path);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700899 if (mounted) {
Al Viro9393bd02009-04-18 13:58:15 -0400900 dput(path->dentry);
901 mntput(path->mnt);
902 path->mnt = mounted;
903 path->dentry = dget(mounted->mnt_root);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700904 return 1;
905 }
906 return 0;
907}
908
Ian Kent62a73752011-03-25 01:51:02 +0800909static inline bool managed_dentry_might_block(struct dentry *dentry)
910{
911 return (dentry->d_flags & DCACHE_MANAGE_TRANSIT &&
912 dentry->d_op->d_manage(dentry, true) < 0);
913}
914
David Howells9875cf82011-01-14 18:45:21 +0000915/*
Al Viro287548e2011-05-27 06:50:06 -0400916 * Try to skip to top of mountpoint pile in rcuwalk mode. Fail if
917 * we meet a managed dentry that would need blocking.
David Howells9875cf82011-01-14 18:45:21 +0000918 */
919static bool __follow_mount_rcu(struct nameidata *nd, struct path *path,
Al Viro287548e2011-05-27 06:50:06 -0400920 struct inode **inode)
David Howells9875cf82011-01-14 18:45:21 +0000921{
Ian Kent62a73752011-03-25 01:51:02 +0800922 for (;;) {
Al Viroc7105362011-11-24 18:22:03 -0500923 struct mount *mounted;
Ian Kent62a73752011-03-25 01:51:02 +0800924 /*
925 * Don't forget we might have a non-mountpoint managed dentry
926 * that wants to block transit.
927 */
Al Viro287548e2011-05-27 06:50:06 -0400928 if (unlikely(managed_dentry_might_block(path->dentry)))
David Howellsab909112011-01-14 18:46:51 +0000929 return false;
Ian Kent62a73752011-03-25 01:51:02 +0800930
931 if (!d_mountpoint(path->dentry))
932 break;
933
David Howells9875cf82011-01-14 18:45:21 +0000934 mounted = __lookup_mnt(path->mnt, path->dentry, 1);
935 if (!mounted)
936 break;
Al Viroc7105362011-11-24 18:22:03 -0500937 path->mnt = &mounted->mnt;
938 path->dentry = mounted->mnt.mnt_root;
Al Viroa3fbbde2011-11-07 21:21:26 +0000939 nd->flags |= LOOKUP_JUMPED;
David Howells9875cf82011-01-14 18:45:21 +0000940 nd->seq = read_seqcount_begin(&path->dentry->d_seq);
Linus Torvalds59430262011-07-18 15:43:29 -0700941 /*
942 * Update the inode too. We don't need to re-check the
943 * dentry sequence number here after this d_inode read,
944 * because a mount-point is always pinned.
945 */
946 *inode = path->dentry->d_inode;
David Howells9875cf82011-01-14 18:45:21 +0000947 }
David Howells9875cf82011-01-14 18:45:21 +0000948 return true;
949}
950
Al Virodea39372011-05-27 06:53:39 -0400951static void follow_mount_rcu(struct nameidata *nd)
Al Viro287548e2011-05-27 06:50:06 -0400952{
Al Virodea39372011-05-27 06:53:39 -0400953 while (d_mountpoint(nd->path.dentry)) {
Al Viroc7105362011-11-24 18:22:03 -0500954 struct mount *mounted;
Al Virodea39372011-05-27 06:53:39 -0400955 mounted = __lookup_mnt(nd->path.mnt, nd->path.dentry, 1);
Al Viro287548e2011-05-27 06:50:06 -0400956 if (!mounted)
957 break;
Al Viroc7105362011-11-24 18:22:03 -0500958 nd->path.mnt = &mounted->mnt;
959 nd->path.dentry = mounted->mnt.mnt_root;
Al Virodea39372011-05-27 06:53:39 -0400960 nd->seq = read_seqcount_begin(&nd->path.dentry->d_seq);
Al Viro287548e2011-05-27 06:50:06 -0400961 }
962}
963
Nick Piggin31e6b012011-01-07 17:49:52 +1100964static int follow_dotdot_rcu(struct nameidata *nd)
965{
Nick Piggin31e6b012011-01-07 17:49:52 +1100966 set_root_rcu(nd);
967
David Howells9875cf82011-01-14 18:45:21 +0000968 while (1) {
Nick Piggin31e6b012011-01-07 17:49:52 +1100969 if (nd->path.dentry == nd->root.dentry &&
970 nd->path.mnt == nd->root.mnt) {
971 break;
972 }
973 if (nd->path.dentry != nd->path.mnt->mnt_root) {
974 struct dentry *old = nd->path.dentry;
975 struct dentry *parent = old->d_parent;
976 unsigned seq;
977
978 seq = read_seqcount_begin(&parent->d_seq);
979 if (read_seqcount_retry(&old->d_seq, nd->seq))
Al Viroef7562d2011-03-04 14:35:59 -0500980 goto failed;
Nick Piggin31e6b012011-01-07 17:49:52 +1100981 nd->path.dentry = parent;
982 nd->seq = seq;
983 break;
984 }
985 if (!follow_up_rcu(&nd->path))
986 break;
987 nd->seq = read_seqcount_begin(&nd->path.dentry->d_seq);
Nick Piggin31e6b012011-01-07 17:49:52 +1100988 }
Al Virodea39372011-05-27 06:53:39 -0400989 follow_mount_rcu(nd);
990 nd->inode = nd->path.dentry->d_inode;
Nick Piggin31e6b012011-01-07 17:49:52 +1100991 return 0;
Al Viroef7562d2011-03-04 14:35:59 -0500992
993failed:
994 nd->flags &= ~LOOKUP_RCU;
Al Viro5b6ca022011-03-09 23:04:47 -0500995 if (!(nd->flags & LOOKUP_ROOT))
996 nd->root.mnt = NULL;
Al Viro32a79912012-07-18 20:43:19 +0400997 unlock_rcu_walk();
Al Viroef7562d2011-03-04 14:35:59 -0500998 return -ECHILD;
Nick Piggin31e6b012011-01-07 17:49:52 +1100999}
1000
David Howells9875cf82011-01-14 18:45:21 +00001001/*
David Howellscc53ce52011-01-14 18:45:26 +00001002 * Follow down to the covering mount currently visible to userspace. At each
1003 * point, the filesystem owning that dentry may be queried as to whether the
1004 * caller is permitted to proceed or not.
David Howellscc53ce52011-01-14 18:45:26 +00001005 */
Al Viro7cc90cc2011-03-18 09:04:20 -04001006int follow_down(struct path *path)
David Howellscc53ce52011-01-14 18:45:26 +00001007{
1008 unsigned managed;
1009 int ret;
1010
1011 while (managed = ACCESS_ONCE(path->dentry->d_flags),
1012 unlikely(managed & DCACHE_MANAGED_DENTRY)) {
1013 /* Allow the filesystem to manage the transit without i_mutex
1014 * being held.
1015 *
1016 * We indicate to the filesystem if someone is trying to mount
1017 * something here. This gives autofs the chance to deny anyone
1018 * other than its daemon the right to mount on its
1019 * superstructure.
1020 *
1021 * The filesystem may sleep at this point.
1022 */
1023 if (managed & DCACHE_MANAGE_TRANSIT) {
1024 BUG_ON(!path->dentry->d_op);
1025 BUG_ON(!path->dentry->d_op->d_manage);
David Howellsab909112011-01-14 18:46:51 +00001026 ret = path->dentry->d_op->d_manage(
Al Viro1aed3e42011-03-18 09:09:02 -04001027 path->dentry, false);
David Howellscc53ce52011-01-14 18:45:26 +00001028 if (ret < 0)
1029 return ret == -EISDIR ? 0 : ret;
1030 }
1031
1032 /* Transit to a mounted filesystem. */
1033 if (managed & DCACHE_MOUNTED) {
1034 struct vfsmount *mounted = lookup_mnt(path);
1035 if (!mounted)
1036 break;
1037 dput(path->dentry);
1038 mntput(path->mnt);
1039 path->mnt = mounted;
1040 path->dentry = dget(mounted->mnt_root);
1041 continue;
1042 }
1043
1044 /* Don't handle automount points here */
1045 break;
1046 }
1047 return 0;
1048}
1049
1050/*
David Howells9875cf82011-01-14 18:45:21 +00001051 * Skip to top of mountpoint pile in refwalk mode for follow_dotdot()
1052 */
1053static void follow_mount(struct path *path)
1054{
1055 while (d_mountpoint(path->dentry)) {
1056 struct vfsmount *mounted = lookup_mnt(path);
1057 if (!mounted)
1058 break;
1059 dput(path->dentry);
1060 mntput(path->mnt);
1061 path->mnt = mounted;
1062 path->dentry = dget(mounted->mnt_root);
1063 }
1064}
1065
Nick Piggin31e6b012011-01-07 17:49:52 +11001066static void follow_dotdot(struct nameidata *nd)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001067{
Al Viro2a737872009-04-07 11:49:53 -04001068 set_root(nd);
Andreas Mohre518ddb2006-09-29 02:01:22 -07001069
Linus Torvalds1da177e2005-04-16 15:20:36 -07001070 while(1) {
Jan Blunck4ac91372008-02-14 19:34:32 -08001071 struct dentry *old = nd->path.dentry;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001072
Al Viro2a737872009-04-07 11:49:53 -04001073 if (nd->path.dentry == nd->root.dentry &&
1074 nd->path.mnt == nd->root.mnt) {
Linus Torvalds1da177e2005-04-16 15:20:36 -07001075 break;
1076 }
Jan Blunck4ac91372008-02-14 19:34:32 -08001077 if (nd->path.dentry != nd->path.mnt->mnt_root) {
Al Viro3088dd72010-01-30 15:47:29 -05001078 /* rare case of legitimate dget_parent()... */
1079 nd->path.dentry = dget_parent(nd->path.dentry);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001080 dput(old);
1081 break;
1082 }
Al Viro3088dd72010-01-30 15:47:29 -05001083 if (!follow_up(&nd->path))
Linus Torvalds1da177e2005-04-16 15:20:36 -07001084 break;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001085 }
Al Viro79ed0222009-04-18 13:59:41 -04001086 follow_mount(&nd->path);
Nick Piggin31e6b012011-01-07 17:49:52 +11001087 nd->inode = nd->path.dentry->d_inode;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001088}
1089
Linus Torvalds1da177e2005-04-16 15:20:36 -07001090/*
Miklos Szeredibad61182012-03-26 12:54:24 +02001091 * This looks up the name in dcache, possibly revalidates the old dentry and
1092 * allocates a new one if not found or not valid. In the need_lookup argument
1093 * returns whether i_op->lookup is necessary.
1094 *
1095 * dir->d_inode->i_mutex must be held
Nick Pigginbaa03892010-08-18 04:37:31 +10001096 */
Miklos Szeredibad61182012-03-26 12:54:24 +02001097static struct dentry *lookup_dcache(struct qstr *name, struct dentry *dir,
Al Viro201f9562012-06-22 12:42:10 +04001098 unsigned int flags, bool *need_lookup)
Nick Pigginbaa03892010-08-18 04:37:31 +10001099{
Nick Pigginbaa03892010-08-18 04:37:31 +10001100 struct dentry *dentry;
Miklos Szeredibad61182012-03-26 12:54:24 +02001101 int error;
Nick Pigginbaa03892010-08-18 04:37:31 +10001102
Miklos Szeredibad61182012-03-26 12:54:24 +02001103 *need_lookup = false;
1104 dentry = d_lookup(dir, name);
1105 if (dentry) {
1106 if (d_need_lookup(dentry)) {
1107 *need_lookup = true;
1108 } else if (dentry->d_flags & DCACHE_OP_REVALIDATE) {
Al Viro201f9562012-06-22 12:42:10 +04001109 error = d_revalidate(dentry, flags);
Miklos Szeredibad61182012-03-26 12:54:24 +02001110 if (unlikely(error <= 0)) {
1111 if (error < 0) {
1112 dput(dentry);
1113 return ERR_PTR(error);
1114 } else if (!d_invalidate(dentry)) {
1115 dput(dentry);
1116 dentry = NULL;
1117 }
1118 }
1119 }
1120 }
Nick Pigginbaa03892010-08-18 04:37:31 +10001121
Miklos Szeredibad61182012-03-26 12:54:24 +02001122 if (!dentry) {
1123 dentry = d_alloc(dir, name);
1124 if (unlikely(!dentry))
1125 return ERR_PTR(-ENOMEM);
Nick Pigginbaa03892010-08-18 04:37:31 +10001126
Miklos Szeredibad61182012-03-26 12:54:24 +02001127 *need_lookup = true;
Nick Pigginbaa03892010-08-18 04:37:31 +10001128 }
1129 return dentry;
1130}
1131
1132/*
Miklos Szeredibad61182012-03-26 12:54:24 +02001133 * Call i_op->lookup on the dentry. The dentry must be negative but may be
1134 * hashed if it was pouplated with DCACHE_NEED_LOOKUP.
1135 *
1136 * dir->d_inode->i_mutex must be held
Josef Bacik44396f42011-05-31 11:58:49 -04001137 */
Miklos Szeredibad61182012-03-26 12:54:24 +02001138static struct dentry *lookup_real(struct inode *dir, struct dentry *dentry,
Al Viro72bd8662012-06-10 17:17:17 -04001139 unsigned int flags)
Josef Bacik44396f42011-05-31 11:58:49 -04001140{
Josef Bacik44396f42011-05-31 11:58:49 -04001141 struct dentry *old;
1142
1143 /* Don't create child dentry for a dead directory. */
Miklos Szeredibad61182012-03-26 12:54:24 +02001144 if (unlikely(IS_DEADDIR(dir))) {
Miklos Szeredie188dc02012-02-03 14:25:18 +01001145 dput(dentry);
Josef Bacik44396f42011-05-31 11:58:49 -04001146 return ERR_PTR(-ENOENT);
Miklos Szeredie188dc02012-02-03 14:25:18 +01001147 }
Josef Bacik44396f42011-05-31 11:58:49 -04001148
Al Viro72bd8662012-06-10 17:17:17 -04001149 old = dir->i_op->lookup(dir, dentry, flags);
Josef Bacik44396f42011-05-31 11:58:49 -04001150 if (unlikely(old)) {
1151 dput(dentry);
1152 dentry = old;
1153 }
1154 return dentry;
1155}
1156
Al Viroa3255542012-03-30 14:41:51 -04001157static struct dentry *__lookup_hash(struct qstr *name,
Al Viro72bd8662012-06-10 17:17:17 -04001158 struct dentry *base, unsigned int flags)
Al Viroa3255542012-03-30 14:41:51 -04001159{
Miklos Szeredibad61182012-03-26 12:54:24 +02001160 bool need_lookup;
Al Viroa3255542012-03-30 14:41:51 -04001161 struct dentry *dentry;
1162
Al Viro72bd8662012-06-10 17:17:17 -04001163 dentry = lookup_dcache(name, base, flags, &need_lookup);
Miklos Szeredibad61182012-03-26 12:54:24 +02001164 if (!need_lookup)
1165 return dentry;
Al Viroa3255542012-03-30 14:41:51 -04001166
Al Viro72bd8662012-06-10 17:17:17 -04001167 return lookup_real(base->d_inode, dentry, flags);
Al Viroa3255542012-03-30 14:41:51 -04001168}
1169
Josef Bacik44396f42011-05-31 11:58:49 -04001170/*
Linus Torvalds1da177e2005-04-16 15:20:36 -07001171 * It's more convoluted than I'd like it to be, but... it's still fairly
1172 * small and for now I'd prefer to have fast path as straight as possible.
1173 * It _is_ time-critical.
1174 */
Miklos Szeredi697f5142012-05-21 17:30:05 +02001175static int lookup_fast(struct nameidata *nd, struct qstr *name,
1176 struct path *path, struct inode **inode)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001177{
Jan Blunck4ac91372008-02-14 19:34:32 -08001178 struct vfsmount *mnt = nd->path.mnt;
Nick Piggin31e6b012011-01-07 17:49:52 +11001179 struct dentry *dentry, *parent = nd->path.dentry;
Al Viro5a18fff2011-03-11 04:44:53 -05001180 int need_reval = 1;
1181 int status = 1;
David Howells9875cf82011-01-14 18:45:21 +00001182 int err;
1183
Al Viro3cac2602009-08-13 18:27:43 +04001184 /*
Nick Pigginb04f7842010-08-18 04:37:34 +10001185 * Rename seqlock is not required here because in the off chance
1186 * of a false negative due to a concurrent rename, we're going to
1187 * do the non-racy lookup, below.
1188 */
Nick Piggin31e6b012011-01-07 17:49:52 +11001189 if (nd->flags & LOOKUP_RCU) {
1190 unsigned seq;
Linus Torvalds12f8ad42012-05-04 14:59:14 -07001191 dentry = __d_lookup_rcu(parent, name, &seq, nd->inode);
Al Viro5a18fff2011-03-11 04:44:53 -05001192 if (!dentry)
1193 goto unlazy;
1194
Linus Torvalds12f8ad42012-05-04 14:59:14 -07001195 /*
1196 * This sequence count validates that the inode matches
1197 * the dentry name information from lookup.
1198 */
1199 *inode = dentry->d_inode;
1200 if (read_seqcount_retry(&dentry->d_seq, seq))
1201 return -ECHILD;
1202
1203 /*
1204 * This sequence count validates that the parent had no
1205 * changes while we did the lookup of the dentry above.
1206 *
1207 * The memory barrier in read_seqcount_begin of child is
1208 * enough, we can use __read_seqcount_retry here.
1209 */
Nick Piggin31e6b012011-01-07 17:49:52 +11001210 if (__read_seqcount_retry(&parent->d_seq, nd->seq))
1211 return -ECHILD;
Nick Piggin31e6b012011-01-07 17:49:52 +11001212 nd->seq = seq;
Al Viro5a18fff2011-03-11 04:44:53 -05001213
Miklos Szeredifa4ee152012-03-26 12:54:19 +02001214 if (unlikely(d_need_lookup(dentry)))
1215 goto unlazy;
Al Viro24643082011-02-15 01:26:22 -05001216 if (unlikely(dentry->d_flags & DCACHE_OP_REVALIDATE)) {
Al Viro4ce16ef32012-06-10 16:10:59 -04001217 status = d_revalidate(dentry, nd->flags);
Al Viro5a18fff2011-03-11 04:44:53 -05001218 if (unlikely(status <= 0)) {
1219 if (status != -ECHILD)
1220 need_reval = 0;
1221 goto unlazy;
1222 }
Al Viro24643082011-02-15 01:26:22 -05001223 }
Nick Piggin31e6b012011-01-07 17:49:52 +11001224 path->mnt = mnt;
1225 path->dentry = dentry;
Al Virod6e9bd22011-05-27 07:03:15 -04001226 if (unlikely(!__follow_mount_rcu(nd, path, inode)))
1227 goto unlazy;
1228 if (unlikely(path->dentry->d_flags & DCACHE_NEED_AUTOMOUNT))
1229 goto unlazy;
1230 return 0;
Al Viro5a18fff2011-03-11 04:44:53 -05001231unlazy:
Al Viro19660af2011-03-25 10:32:48 -04001232 if (unlazy_walk(nd, dentry))
1233 return -ECHILD;
Al Viro5a18fff2011-03-11 04:44:53 -05001234 } else {
1235 dentry = __d_lookup(parent, name);
Nick Piggin31e6b012011-01-07 17:49:52 +11001236 }
Al Viro5a18fff2011-03-11 04:44:53 -05001237
Al Viro81e6f522012-03-30 14:48:04 -04001238 if (unlikely(!dentry))
1239 goto need_lookup;
Al Viro5a18fff2011-03-11 04:44:53 -05001240
Al Viro81e6f522012-03-30 14:48:04 -04001241 if (unlikely(d_need_lookup(dentry))) {
1242 dput(dentry);
1243 goto need_lookup;
Al Viro24643082011-02-15 01:26:22 -05001244 }
Al Viro81e6f522012-03-30 14:48:04 -04001245
Al Viro5a18fff2011-03-11 04:44:53 -05001246 if (unlikely(dentry->d_flags & DCACHE_OP_REVALIDATE) && need_reval)
Al Viro4ce16ef32012-06-10 16:10:59 -04001247 status = d_revalidate(dentry, nd->flags);
Al Viro5a18fff2011-03-11 04:44:53 -05001248 if (unlikely(status <= 0)) {
1249 if (status < 0) {
1250 dput(dentry);
1251 return status;
1252 }
1253 if (!d_invalidate(dentry)) {
1254 dput(dentry);
Al Viro81e6f522012-03-30 14:48:04 -04001255 goto need_lookup;
Al Viro5a18fff2011-03-11 04:44:53 -05001256 }
1257 }
Miklos Szeredi697f5142012-05-21 17:30:05 +02001258
David Howells9875cf82011-01-14 18:45:21 +00001259 path->mnt = mnt;
1260 path->dentry = dentry;
1261 err = follow_managed(path, nd->flags);
Ian Kent89312212011-01-18 12:06:10 +08001262 if (unlikely(err < 0)) {
1263 path_put_conditional(path, nd);
David Howells9875cf82011-01-14 18:45:21 +00001264 return err;
Ian Kent89312212011-01-18 12:06:10 +08001265 }
Al Viroa3fbbde2011-11-07 21:21:26 +00001266 if (err)
1267 nd->flags |= LOOKUP_JUMPED;
David Howells9875cf82011-01-14 18:45:21 +00001268 *inode = path->dentry->d_inode;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001269 return 0;
Al Viro81e6f522012-03-30 14:48:04 -04001270
1271need_lookup:
Miklos Szeredi697f5142012-05-21 17:30:05 +02001272 return 1;
1273}
1274
1275/* Fast lookup failed, do it the slow way */
1276static int lookup_slow(struct nameidata *nd, struct qstr *name,
1277 struct path *path)
1278{
1279 struct dentry *dentry, *parent;
1280 int err;
1281
1282 parent = nd->path.dentry;
Al Viro81e6f522012-03-30 14:48:04 -04001283 BUG_ON(nd->inode != parent->d_inode);
1284
1285 mutex_lock(&parent->d_inode->i_mutex);
Al Viro72bd8662012-06-10 17:17:17 -04001286 dentry = __lookup_hash(name, parent, nd->flags);
Al Viro81e6f522012-03-30 14:48:04 -04001287 mutex_unlock(&parent->d_inode->i_mutex);
1288 if (IS_ERR(dentry))
1289 return PTR_ERR(dentry);
Miklos Szeredi697f5142012-05-21 17:30:05 +02001290 path->mnt = nd->path.mnt;
1291 path->dentry = dentry;
1292 err = follow_managed(path, nd->flags);
1293 if (unlikely(err < 0)) {
1294 path_put_conditional(path, nd);
1295 return err;
1296 }
1297 if (err)
1298 nd->flags |= LOOKUP_JUMPED;
1299 return 0;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001300}
1301
Al Viro52094c82011-02-21 21:34:47 -05001302static inline int may_lookup(struct nameidata *nd)
1303{
1304 if (nd->flags & LOOKUP_RCU) {
Al Viro4ad5abb2011-06-20 19:57:03 -04001305 int err = inode_permission(nd->inode, MAY_EXEC|MAY_NOT_BLOCK);
Al Viro52094c82011-02-21 21:34:47 -05001306 if (err != -ECHILD)
1307 return err;
Al Viro19660af2011-03-25 10:32:48 -04001308 if (unlazy_walk(nd, NULL))
Al Viro52094c82011-02-21 21:34:47 -05001309 return -ECHILD;
1310 }
Al Viro4ad5abb2011-06-20 19:57:03 -04001311 return inode_permission(nd->inode, MAY_EXEC);
Al Viro52094c82011-02-21 21:34:47 -05001312}
1313
Al Viro9856fa12011-03-04 14:22:06 -05001314static inline int handle_dots(struct nameidata *nd, int type)
1315{
1316 if (type == LAST_DOTDOT) {
1317 if (nd->flags & LOOKUP_RCU) {
1318 if (follow_dotdot_rcu(nd))
1319 return -ECHILD;
1320 } else
1321 follow_dotdot(nd);
1322 }
1323 return 0;
1324}
1325
Al Viro951361f2011-03-04 14:44:37 -05001326static void terminate_walk(struct nameidata *nd)
1327{
1328 if (!(nd->flags & LOOKUP_RCU)) {
1329 path_put(&nd->path);
1330 } else {
1331 nd->flags &= ~LOOKUP_RCU;
Al Viro5b6ca022011-03-09 23:04:47 -05001332 if (!(nd->flags & LOOKUP_ROOT))
1333 nd->root.mnt = NULL;
Al Viro32a79912012-07-18 20:43:19 +04001334 unlock_rcu_walk();
Al Viro951361f2011-03-04 14:44:37 -05001335 }
1336}
1337
Linus Torvalds3ddcd052011-08-06 22:45:50 -07001338/*
1339 * Do we need to follow links? We _really_ want to be able
1340 * to do this check without having to look at inode->i_op,
1341 * so we keep a cache of "no, this doesn't need follow_link"
1342 * for the common case.
1343 */
Linus Torvalds7813b942011-08-07 09:53:20 -07001344static inline int should_follow_link(struct inode *inode, int follow)
Linus Torvalds3ddcd052011-08-06 22:45:50 -07001345{
1346 if (unlikely(!(inode->i_opflags & IOP_NOFOLLOW))) {
1347 if (likely(inode->i_op->follow_link))
1348 return follow;
1349
1350 /* This gets set once for the inode lifetime */
1351 spin_lock(&inode->i_lock);
1352 inode->i_opflags |= IOP_NOFOLLOW;
1353 spin_unlock(&inode->i_lock);
1354 }
1355 return 0;
1356}
1357
Al Viroce57dfc2011-03-13 19:58:58 -04001358static inline int walk_component(struct nameidata *nd, struct path *path,
1359 struct qstr *name, int type, int follow)
1360{
1361 struct inode *inode;
1362 int err;
1363 /*
1364 * "." and ".." are special - ".." especially so because it has
1365 * to be able to know about the current root directory and
1366 * parent relationships.
1367 */
1368 if (unlikely(type != LAST_NORM))
1369 return handle_dots(nd, type);
Miklos Szeredi697f5142012-05-21 17:30:05 +02001370 err = lookup_fast(nd, name, path, &inode);
Al Viroce57dfc2011-03-13 19:58:58 -04001371 if (unlikely(err)) {
Miklos Szeredi697f5142012-05-21 17:30:05 +02001372 if (err < 0)
1373 goto out_err;
1374
1375 err = lookup_slow(nd, name, path);
1376 if (err < 0)
1377 goto out_err;
1378
1379 inode = path->dentry->d_inode;
Al Viroce57dfc2011-03-13 19:58:58 -04001380 }
Miklos Szeredi697f5142012-05-21 17:30:05 +02001381 err = -ENOENT;
1382 if (!inode)
1383 goto out_path_put;
1384
Linus Torvalds7813b942011-08-07 09:53:20 -07001385 if (should_follow_link(inode, follow)) {
Al Viro19660af2011-03-25 10:32:48 -04001386 if (nd->flags & LOOKUP_RCU) {
1387 if (unlikely(unlazy_walk(nd, path->dentry))) {
Miklos Szeredi697f5142012-05-21 17:30:05 +02001388 err = -ECHILD;
1389 goto out_err;
Al Viro19660af2011-03-25 10:32:48 -04001390 }
1391 }
Al Viroce57dfc2011-03-13 19:58:58 -04001392 BUG_ON(inode != path->dentry->d_inode);
1393 return 1;
1394 }
1395 path_to_nameidata(path, nd);
1396 nd->inode = inode;
1397 return 0;
Miklos Szeredi697f5142012-05-21 17:30:05 +02001398
1399out_path_put:
1400 path_to_nameidata(path, nd);
1401out_err:
1402 terminate_walk(nd);
1403 return err;
Al Viroce57dfc2011-03-13 19:58:58 -04001404}
1405
Linus Torvalds1da177e2005-04-16 15:20:36 -07001406/*
Al Virob3563792011-03-14 21:54:55 -04001407 * This limits recursive symlink follows to 8, while
1408 * limiting consecutive symlinks to 40.
1409 *
1410 * Without that kind of total limit, nasty chains of consecutive
1411 * symlinks can cause almost arbitrarily long lookups.
1412 */
1413static inline int nested_symlink(struct path *path, struct nameidata *nd)
1414{
1415 int res;
1416
Al Virob3563792011-03-14 21:54:55 -04001417 if (unlikely(current->link_count >= MAX_NESTED_LINKS)) {
1418 path_put_conditional(path, nd);
1419 path_put(&nd->path);
1420 return -ELOOP;
1421 }
Erez Zadok1a4022f2011-05-21 01:19:59 -04001422 BUG_ON(nd->depth >= MAX_NESTED_LINKS);
Al Virob3563792011-03-14 21:54:55 -04001423
1424 nd->depth++;
1425 current->link_count++;
1426
1427 do {
1428 struct path link = *path;
1429 void *cookie;
Al Viro574197e2011-03-14 22:20:34 -04001430
1431 res = follow_link(&link, nd, &cookie);
Al Viro6d7b5aa2012-06-10 04:15:17 -04001432 if (res)
1433 break;
1434 res = walk_component(nd, path, &nd->last,
1435 nd->last_type, LOOKUP_FOLLOW);
Al Viro574197e2011-03-14 22:20:34 -04001436 put_link(nd, &link, cookie);
Al Virob3563792011-03-14 21:54:55 -04001437 } while (res > 0);
1438
1439 current->link_count--;
1440 nd->depth--;
1441 return res;
1442}
1443
1444/*
Linus Torvalds3ddcd052011-08-06 22:45:50 -07001445 * We really don't want to look at inode->i_op->lookup
1446 * when we don't have to. So we keep a cache bit in
1447 * the inode ->i_opflags field that says "yes, we can
1448 * do lookup on this inode".
1449 */
1450static inline int can_lookup(struct inode *inode)
1451{
1452 if (likely(inode->i_opflags & IOP_LOOKUP))
1453 return 1;
1454 if (likely(!inode->i_op->lookup))
1455 return 0;
1456
1457 /* We do this once for the lifetime of the inode */
1458 spin_lock(&inode->i_lock);
1459 inode->i_opflags |= IOP_LOOKUP;
1460 spin_unlock(&inode->i_lock);
1461 return 1;
1462}
1463
Linus Torvaldsbfcfaa72012-03-06 11:16:17 -08001464/*
1465 * We can do the critical dentry name comparison and hashing
1466 * operations one word at a time, but we are limited to:
1467 *
1468 * - Architectures with fast unaligned word accesses. We could
1469 * do a "get_unaligned()" if this helps and is sufficiently
1470 * fast.
1471 *
1472 * - Little-endian machines (so that we can generate the mask
1473 * of low bytes efficiently). Again, we *could* do a byte
1474 * swapping load on big-endian architectures if that is not
1475 * expensive enough to make the optimization worthless.
1476 *
1477 * - non-CONFIG_DEBUG_PAGEALLOC configurations (so that we
1478 * do not trap on the (extremely unlikely) case of a page
1479 * crossing operation.
1480 *
1481 * - Furthermore, we need an efficient 64-bit compile for the
1482 * 64-bit case in order to generate the "number of bytes in
1483 * the final mask". Again, that could be replaced with a
1484 * efficient population count instruction or similar.
1485 */
1486#ifdef CONFIG_DCACHE_WORD_ACCESS
1487
Linus Torvaldsf68e5562012-04-06 13:54:56 -07001488#include <asm/word-at-a-time.h>
Linus Torvaldsbfcfaa72012-03-06 11:16:17 -08001489
Linus Torvaldsf68e5562012-04-06 13:54:56 -07001490#ifdef CONFIG_64BIT
Linus Torvaldsbfcfaa72012-03-06 11:16:17 -08001491
1492static inline unsigned int fold_hash(unsigned long hash)
1493{
1494 hash += hash >> (8*sizeof(int));
1495 return hash;
1496}
1497
1498#else /* 32-bit case */
1499
Linus Torvaldsbfcfaa72012-03-06 11:16:17 -08001500#define fold_hash(x) (x)
1501
1502#endif
1503
1504unsigned int full_name_hash(const unsigned char *name, unsigned int len)
1505{
1506 unsigned long a, mask;
1507 unsigned long hash = 0;
1508
1509 for (;;) {
Linus Torvaldse419b4c2012-05-03 10:16:43 -07001510 a = load_unaligned_zeropad(name);
Linus Torvaldsbfcfaa72012-03-06 11:16:17 -08001511 if (len < sizeof(unsigned long))
1512 break;
1513 hash += a;
Al Virof132c5b2012-03-22 21:59:52 +00001514 hash *= 9;
Linus Torvaldsbfcfaa72012-03-06 11:16:17 -08001515 name += sizeof(unsigned long);
1516 len -= sizeof(unsigned long);
1517 if (!len)
1518 goto done;
1519 }
1520 mask = ~(~0ul << len*8);
1521 hash += mask & a;
1522done:
1523 return fold_hash(hash);
1524}
1525EXPORT_SYMBOL(full_name_hash);
1526
Linus Torvaldsbfcfaa72012-03-06 11:16:17 -08001527/*
1528 * Calculate the length and hash of the path component, and
1529 * return the length of the component;
1530 */
1531static inline unsigned long hash_name(const char *name, unsigned int *hashp)
1532{
Linus Torvalds36126f82012-05-26 10:43:17 -07001533 unsigned long a, b, adata, bdata, mask, hash, len;
1534 const struct word_at_a_time constants = WORD_AT_A_TIME_CONSTANTS;
Linus Torvaldsbfcfaa72012-03-06 11:16:17 -08001535
1536 hash = a = 0;
1537 len = -sizeof(unsigned long);
1538 do {
1539 hash = (hash + a) * 9;
1540 len += sizeof(unsigned long);
Linus Torvaldse419b4c2012-05-03 10:16:43 -07001541 a = load_unaligned_zeropad(name+len);
Linus Torvalds36126f82012-05-26 10:43:17 -07001542 b = a ^ REPEAT_BYTE('/');
1543 } while (!(has_zero(a, &adata, &constants) | has_zero(b, &bdata, &constants)));
Linus Torvaldsbfcfaa72012-03-06 11:16:17 -08001544
Linus Torvalds36126f82012-05-26 10:43:17 -07001545 adata = prep_zero_mask(a, adata, &constants);
1546 bdata = prep_zero_mask(b, bdata, &constants);
1547
1548 mask = create_zero_mask(adata | bdata);
1549
1550 hash += a & zero_bytemask(mask);
Linus Torvaldsbfcfaa72012-03-06 11:16:17 -08001551 *hashp = fold_hash(hash);
1552
Linus Torvalds36126f82012-05-26 10:43:17 -07001553 return len + find_zero(mask);
Linus Torvaldsbfcfaa72012-03-06 11:16:17 -08001554}
1555
1556#else
1557
Linus Torvalds0145acc2012-03-02 14:32:59 -08001558unsigned int full_name_hash(const unsigned char *name, unsigned int len)
1559{
1560 unsigned long hash = init_name_hash();
1561 while (len--)
1562 hash = partial_name_hash(*name++, hash);
1563 return end_name_hash(hash);
1564}
Linus Torvaldsae942ae2012-03-02 19:40:57 -08001565EXPORT_SYMBOL(full_name_hash);
Linus Torvalds0145acc2012-03-02 14:32:59 -08001566
Linus Torvalds3ddcd052011-08-06 22:45:50 -07001567/*
Linus Torvalds200e9ef2012-03-02 14:49:24 -08001568 * We know there's a real path component here of at least
1569 * one character.
1570 */
1571static inline unsigned long hash_name(const char *name, unsigned int *hashp)
1572{
1573 unsigned long hash = init_name_hash();
1574 unsigned long len = 0, c;
1575
1576 c = (unsigned char)*name;
1577 do {
1578 len++;
1579 hash = partial_name_hash(c, hash);
1580 c = (unsigned char)name[len];
1581 } while (c && c != '/');
1582 *hashp = end_name_hash(hash);
1583 return len;
1584}
1585
Linus Torvaldsbfcfaa72012-03-06 11:16:17 -08001586#endif
1587
Linus Torvalds200e9ef2012-03-02 14:49:24 -08001588/*
Linus Torvalds1da177e2005-04-16 15:20:36 -07001589 * Name resolution.
Prasanna Medaea3834d2005-04-29 16:00:17 +01001590 * This is the basic name resolution function, turning a pathname into
1591 * the final dentry. We expect 'base' to be positive and a directory.
Linus Torvalds1da177e2005-04-16 15:20:36 -07001592 *
Prasanna Medaea3834d2005-04-29 16:00:17 +01001593 * Returns 0 and nd will have valid dentry and mnt on success.
1594 * Returns error and drops reference to input namei data on failure.
Linus Torvalds1da177e2005-04-16 15:20:36 -07001595 */
Al Viro6de88d72009-08-09 01:41:57 +04001596static int link_path_walk(const char *name, struct nameidata *nd)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001597{
1598 struct path next;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001599 int err;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001600
1601 while (*name=='/')
1602 name++;
1603 if (!*name)
Al Viro086e1832011-02-22 20:56:27 -05001604 return 0;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001605
Linus Torvalds1da177e2005-04-16 15:20:36 -07001606 /* At this point we know we have a real path component. */
1607 for(;;) {
Linus Torvalds1da177e2005-04-16 15:20:36 -07001608 struct qstr this;
Linus Torvalds200e9ef2012-03-02 14:49:24 -08001609 long len;
Al Virofe479a52011-02-22 15:10:03 -05001610 int type;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001611
Al Viro52094c82011-02-21 21:34:47 -05001612 err = may_lookup(nd);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001613 if (err)
1614 break;
1615
Linus Torvalds200e9ef2012-03-02 14:49:24 -08001616 len = hash_name(name, &this.hash);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001617 this.name = name;
Linus Torvalds200e9ef2012-03-02 14:49:24 -08001618 this.len = len;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001619
Al Virofe479a52011-02-22 15:10:03 -05001620 type = LAST_NORM;
Linus Torvalds200e9ef2012-03-02 14:49:24 -08001621 if (name[0] == '.') switch (len) {
Al Virofe479a52011-02-22 15:10:03 -05001622 case 2:
Linus Torvalds200e9ef2012-03-02 14:49:24 -08001623 if (name[1] == '.') {
Al Virofe479a52011-02-22 15:10:03 -05001624 type = LAST_DOTDOT;
Al Viro16c2cd72011-02-22 15:50:10 -05001625 nd->flags |= LOOKUP_JUMPED;
1626 }
Al Virofe479a52011-02-22 15:10:03 -05001627 break;
1628 case 1:
1629 type = LAST_DOT;
1630 }
Al Viro5a202bc2011-03-08 14:17:44 -05001631 if (likely(type == LAST_NORM)) {
1632 struct dentry *parent = nd->path.dentry;
Al Viro16c2cd72011-02-22 15:50:10 -05001633 nd->flags &= ~LOOKUP_JUMPED;
Al Viro5a202bc2011-03-08 14:17:44 -05001634 if (unlikely(parent->d_flags & DCACHE_OP_HASH)) {
1635 err = parent->d_op->d_hash(parent, nd->inode,
1636 &this);
1637 if (err < 0)
1638 break;
1639 }
1640 }
Al Virofe479a52011-02-22 15:10:03 -05001641
Linus Torvalds200e9ef2012-03-02 14:49:24 -08001642 if (!name[len])
Linus Torvalds1da177e2005-04-16 15:20:36 -07001643 goto last_component;
Linus Torvalds200e9ef2012-03-02 14:49:24 -08001644 /*
1645 * If it wasn't NUL, we know it was '/'. Skip that
1646 * slash, and continue until no more slashes.
1647 */
1648 do {
1649 len++;
1650 } while (unlikely(name[len] == '/'));
1651 if (!name[len])
Al Virob3563792011-03-14 21:54:55 -04001652 goto last_component;
Linus Torvalds200e9ef2012-03-02 14:49:24 -08001653 name += len;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001654
Al Viroce57dfc2011-03-13 19:58:58 -04001655 err = walk_component(nd, &next, &this, type, LOOKUP_FOLLOW);
1656 if (err < 0)
1657 return err;
Al Virofe479a52011-02-22 15:10:03 -05001658
Al Viroce57dfc2011-03-13 19:58:58 -04001659 if (err) {
Al Virob3563792011-03-14 21:54:55 -04001660 err = nested_symlink(&next, nd);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001661 if (err)
Al Viroa7472ba2011-03-04 14:39:30 -05001662 return err;
Nick Piggin31e6b012011-01-07 17:49:52 +11001663 }
Linus Torvalds3ddcd052011-08-06 22:45:50 -07001664 if (can_lookup(nd->inode))
1665 continue;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001666 err = -ENOTDIR;
Linus Torvalds3ddcd052011-08-06 22:45:50 -07001667 break;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001668 /* here ends the main loop */
1669
Linus Torvalds1da177e2005-04-16 15:20:36 -07001670last_component:
Al Virob3563792011-03-14 21:54:55 -04001671 nd->last = this;
1672 nd->last_type = type;
Al Viro086e1832011-02-22 20:56:27 -05001673 return 0;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001674 }
Al Viro951361f2011-03-04 14:44:37 -05001675 terminate_walk(nd);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001676 return err;
1677}
1678
Al Viro70e9b352011-03-05 21:12:22 -05001679static int path_init(int dfd, const char *name, unsigned int flags,
1680 struct nameidata *nd, struct file **fp)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001681{
Prasanna Medaea3834d2005-04-29 16:00:17 +01001682 int retval = 0;
Ulrich Drepper170aa3d2006-02-04 23:28:02 -08001683 int fput_needed;
1684 struct file *file;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001685
1686 nd->last_type = LAST_ROOT; /* if there are only slashes... */
Al Viro16c2cd72011-02-22 15:50:10 -05001687 nd->flags = flags | LOOKUP_JUMPED;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001688 nd->depth = 0;
Al Viro5b6ca022011-03-09 23:04:47 -05001689 if (flags & LOOKUP_ROOT) {
1690 struct inode *inode = nd->root.dentry->d_inode;
Al Viro73d049a2011-03-11 12:08:24 -05001691 if (*name) {
1692 if (!inode->i_op->lookup)
1693 return -ENOTDIR;
1694 retval = inode_permission(inode, MAY_EXEC);
1695 if (retval)
1696 return retval;
1697 }
Al Viro5b6ca022011-03-09 23:04:47 -05001698 nd->path = nd->root;
1699 nd->inode = inode;
1700 if (flags & LOOKUP_RCU) {
Al Viro32a79912012-07-18 20:43:19 +04001701 lock_rcu_walk();
Al Viro5b6ca022011-03-09 23:04:47 -05001702 nd->seq = __read_seqcount_begin(&nd->path.dentry->d_seq);
1703 } else {
1704 path_get(&nd->path);
1705 }
1706 return 0;
1707 }
1708
Al Viro2a737872009-04-07 11:49:53 -04001709 nd->root.mnt = NULL;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001710
Linus Torvalds1da177e2005-04-16 15:20:36 -07001711 if (*name=='/') {
Al Viroe41f7d42011-02-22 14:02:58 -05001712 if (flags & LOOKUP_RCU) {
Al Viro32a79912012-07-18 20:43:19 +04001713 lock_rcu_walk();
Al Viroe41f7d42011-02-22 14:02:58 -05001714 set_root_rcu(nd);
1715 } else {
1716 set_root(nd);
1717 path_get(&nd->root);
1718 }
Al Viro2a737872009-04-07 11:49:53 -04001719 nd->path = nd->root;
Ulrich Drepper5590ff02006-01-18 17:43:53 -08001720 } else if (dfd == AT_FDCWD) {
Al Viroe41f7d42011-02-22 14:02:58 -05001721 if (flags & LOOKUP_RCU) {
1722 struct fs_struct *fs = current->fs;
1723 unsigned seq;
1724
Al Viro32a79912012-07-18 20:43:19 +04001725 lock_rcu_walk();
Al Viroe41f7d42011-02-22 14:02:58 -05001726
1727 do {
1728 seq = read_seqcount_begin(&fs->seq);
1729 nd->path = fs->pwd;
1730 nd->seq = __read_seqcount_begin(&nd->path.dentry->d_seq);
1731 } while (read_seqcount_retry(&fs->seq, seq));
1732 } else {
1733 get_fs_pwd(current->fs, &nd->path);
1734 }
Ulrich Drepper5590ff02006-01-18 17:43:53 -08001735 } else {
Ulrich Drepper5590ff02006-01-18 17:43:53 -08001736 struct dentry *dentry;
1737
Al Viro1abf0c72011-03-13 03:51:11 -04001738 file = fget_raw_light(dfd, &fput_needed);
Ulrich Drepper170aa3d2006-02-04 23:28:02 -08001739 retval = -EBADF;
1740 if (!file)
Trond Myklebust6d09bb62006-06-04 02:51:37 -07001741 goto out_fail;
Ulrich Drepper5590ff02006-01-18 17:43:53 -08001742
Josef "Jeff" Sipek0f7fc9e2006-12-08 02:36:35 -08001743 dentry = file->f_path.dentry;
Ulrich Drepper5590ff02006-01-18 17:43:53 -08001744
Al Virof52e0c12011-03-14 18:56:51 -04001745 if (*name) {
1746 retval = -ENOTDIR;
1747 if (!S_ISDIR(dentry->d_inode->i_mode))
1748 goto fput_fail;
Ulrich Drepper5590ff02006-01-18 17:43:53 -08001749
Al Viro4ad5abb2011-06-20 19:57:03 -04001750 retval = inode_permission(dentry->d_inode, MAY_EXEC);
Al Virof52e0c12011-03-14 18:56:51 -04001751 if (retval)
1752 goto fput_fail;
1753 }
Ulrich Drepper5590ff02006-01-18 17:43:53 -08001754
Jan Blunck5dd784d02008-02-14 19:34:38 -08001755 nd->path = file->f_path;
Al Viroe41f7d42011-02-22 14:02:58 -05001756 if (flags & LOOKUP_RCU) {
1757 if (fput_needed)
Al Viro70e9b352011-03-05 21:12:22 -05001758 *fp = file;
Al Viroe41f7d42011-02-22 14:02:58 -05001759 nd->seq = __read_seqcount_begin(&nd->path.dentry->d_seq);
Al Viro32a79912012-07-18 20:43:19 +04001760 lock_rcu_walk();
Al Viroe41f7d42011-02-22 14:02:58 -05001761 } else {
1762 path_get(&file->f_path);
1763 fput_light(file, fput_needed);
1764 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07001765 }
Al Viroe41f7d42011-02-22 14:02:58 -05001766
Nick Piggin31e6b012011-01-07 17:49:52 +11001767 nd->inode = nd->path.dentry->d_inode;
Al Viro9b4a9b12009-04-07 11:44:16 -04001768 return 0;
Ulrich Drepper170aa3d2006-02-04 23:28:02 -08001769
Trond Myklebust6d09bb62006-06-04 02:51:37 -07001770fput_fail:
Ulrich Drepper170aa3d2006-02-04 23:28:02 -08001771 fput_light(file, fput_needed);
Al Viro9b4a9b12009-04-07 11:44:16 -04001772out_fail:
1773 return retval;
1774}
1775
Al Virobd92d7f2011-03-14 19:54:59 -04001776static inline int lookup_last(struct nameidata *nd, struct path *path)
1777{
1778 if (nd->last_type == LAST_NORM && nd->last.name[nd->last.len])
1779 nd->flags |= LOOKUP_FOLLOW | LOOKUP_DIRECTORY;
1780
1781 nd->flags &= ~LOOKUP_PARENT;
1782 return walk_component(nd, path, &nd->last, nd->last_type,
1783 nd->flags & LOOKUP_FOLLOW);
1784}
1785
Al Viro9b4a9b12009-04-07 11:44:16 -04001786/* Returns 0 and nd will be valid on success; Retuns error, otherwise. */
Al Viroee0827c2011-02-21 23:38:09 -05001787static int path_lookupat(int dfd, const char *name,
Al Viro9b4a9b12009-04-07 11:44:16 -04001788 unsigned int flags, struct nameidata *nd)
1789{
Al Viro70e9b352011-03-05 21:12:22 -05001790 struct file *base = NULL;
Al Virobd92d7f2011-03-14 19:54:59 -04001791 struct path path;
1792 int err;
Nick Piggin31e6b012011-01-07 17:49:52 +11001793
1794 /*
1795 * Path walking is largely split up into 2 different synchronisation
1796 * schemes, rcu-walk and ref-walk (explained in
1797 * Documentation/filesystems/path-lookup.txt). These share much of the
1798 * path walk code, but some things particularly setup, cleanup, and
1799 * following mounts are sufficiently divergent that functions are
1800 * duplicated. Typically there is a function foo(), and its RCU
1801 * analogue, foo_rcu().
1802 *
1803 * -ECHILD is the error number of choice (just to avoid clashes) that
1804 * is returned if some aspect of an rcu-walk fails. Such an error must
1805 * be handled by restarting a traditional ref-walk (which will always
1806 * be able to complete).
1807 */
Al Virobd92d7f2011-03-14 19:54:59 -04001808 err = path_init(dfd, name, flags | LOOKUP_PARENT, nd, &base);
Al Viroee0827c2011-02-21 23:38:09 -05001809
Al Virobd92d7f2011-03-14 19:54:59 -04001810 if (unlikely(err))
1811 return err;
Al Viroee0827c2011-02-21 23:38:09 -05001812
1813 current->total_link_count = 0;
Al Virobd92d7f2011-03-14 19:54:59 -04001814 err = link_path_walk(name, nd);
1815
1816 if (!err && !(flags & LOOKUP_PARENT)) {
Al Virobd92d7f2011-03-14 19:54:59 -04001817 err = lookup_last(nd, &path);
1818 while (err > 0) {
1819 void *cookie;
1820 struct path link = path;
Al Virobd92d7f2011-03-14 19:54:59 -04001821 nd->flags |= LOOKUP_PARENT;
Al Viro574197e2011-03-14 22:20:34 -04001822 err = follow_link(&link, nd, &cookie);
Al Viro6d7b5aa2012-06-10 04:15:17 -04001823 if (err)
1824 break;
1825 err = lookup_last(nd, &path);
Al Viro574197e2011-03-14 22:20:34 -04001826 put_link(nd, &link, cookie);
Al Virobd92d7f2011-03-14 19:54:59 -04001827 }
1828 }
Al Viroee0827c2011-02-21 23:38:09 -05001829
Al Viro9f1fafe2011-03-25 11:00:12 -04001830 if (!err)
1831 err = complete_walk(nd);
Al Virobd92d7f2011-03-14 19:54:59 -04001832
1833 if (!err && nd->flags & LOOKUP_DIRECTORY) {
1834 if (!nd->inode->i_op->lookup) {
1835 path_put(&nd->path);
Al Virobd23a532011-03-23 09:56:30 -04001836 err = -ENOTDIR;
Al Virobd92d7f2011-03-14 19:54:59 -04001837 }
1838 }
Al Viro16c2cd72011-02-22 15:50:10 -05001839
Al Viro70e9b352011-03-05 21:12:22 -05001840 if (base)
1841 fput(base);
Al Viroee0827c2011-02-21 23:38:09 -05001842
Al Viro5b6ca022011-03-09 23:04:47 -05001843 if (nd->root.mnt && !(nd->flags & LOOKUP_ROOT)) {
Al Viro2a737872009-04-07 11:49:53 -04001844 path_put(&nd->root);
1845 nd->root.mnt = NULL;
1846 }
Al Virobd92d7f2011-03-14 19:54:59 -04001847 return err;
Al Viroee0827c2011-02-21 23:38:09 -05001848}
Nick Piggin31e6b012011-01-07 17:49:52 +11001849
Al Viroee0827c2011-02-21 23:38:09 -05001850static int do_path_lookup(int dfd, const char *name,
1851 unsigned int flags, struct nameidata *nd)
1852{
1853 int retval = path_lookupat(dfd, name, flags | LOOKUP_RCU, nd);
1854 if (unlikely(retval == -ECHILD))
1855 retval = path_lookupat(dfd, name, flags, nd);
1856 if (unlikely(retval == -ESTALE))
1857 retval = path_lookupat(dfd, name, flags | LOOKUP_REVAL, nd);
Nick Piggin31e6b012011-01-07 17:49:52 +11001858
1859 if (likely(!retval)) {
1860 if (unlikely(!audit_dummy_context())) {
1861 if (nd->path.dentry && nd->inode)
1862 audit_inode(name, nd->path.dentry);
1863 }
1864 }
Al Viro9b4a9b12009-04-07 11:44:16 -04001865 return retval;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001866}
1867
Al Viro79714f72012-06-15 03:01:42 +04001868/* does lookup, returns the object with parent locked */
1869struct dentry *kern_path_locked(const char *name, struct path *path)
Ulrich Drepper5590ff02006-01-18 17:43:53 -08001870{
Al Viro79714f72012-06-15 03:01:42 +04001871 struct nameidata nd;
1872 struct dentry *d;
1873 int err = do_path_lookup(AT_FDCWD, name, LOOKUP_PARENT, &nd);
1874 if (err)
1875 return ERR_PTR(err);
1876 if (nd.last_type != LAST_NORM) {
1877 path_put(&nd.path);
1878 return ERR_PTR(-EINVAL);
1879 }
1880 mutex_lock_nested(&nd.path.dentry->d_inode->i_mutex, I_MUTEX_PARENT);
Al Viro1e0ea002012-07-22 23:46:21 +04001881 d = __lookup_hash(&nd.last, nd.path.dentry, 0);
Al Viro79714f72012-06-15 03:01:42 +04001882 if (IS_ERR(d)) {
1883 mutex_unlock(&nd.path.dentry->d_inode->i_mutex);
1884 path_put(&nd.path);
1885 return d;
1886 }
1887 *path = nd.path;
1888 return d;
Ulrich Drepper5590ff02006-01-18 17:43:53 -08001889}
1890
Al Virod1811462008-08-02 00:49:18 -04001891int kern_path(const char *name, unsigned int flags, struct path *path)
1892{
1893 struct nameidata nd;
1894 int res = do_path_lookup(AT_FDCWD, name, flags, &nd);
1895 if (!res)
1896 *path = nd.path;
1897 return res;
1898}
1899
Josef 'Jeff' Sipek16f18202007-07-19 01:48:18 -07001900/**
1901 * vfs_path_lookup - lookup a file path relative to a dentry-vfsmount pair
1902 * @dentry: pointer to dentry of the base directory
1903 * @mnt: pointer to vfs mount of the base directory
1904 * @name: pointer to file name
1905 * @flags: lookup flags
Al Viroe0a01242011-06-27 17:00:37 -04001906 * @path: pointer to struct path to fill
Josef 'Jeff' Sipek16f18202007-07-19 01:48:18 -07001907 */
1908int vfs_path_lookup(struct dentry *dentry, struct vfsmount *mnt,
1909 const char *name, unsigned int flags,
Al Viroe0a01242011-06-27 17:00:37 -04001910 struct path *path)
Josef 'Jeff' Sipek16f18202007-07-19 01:48:18 -07001911{
Al Viroe0a01242011-06-27 17:00:37 -04001912 struct nameidata nd;
1913 int err;
1914 nd.root.dentry = dentry;
1915 nd.root.mnt = mnt;
1916 BUG_ON(flags & LOOKUP_PARENT);
Al Viro5b6ca022011-03-09 23:04:47 -05001917 /* the first argument of do_path_lookup() is ignored with LOOKUP_ROOT */
Al Viroe0a01242011-06-27 17:00:37 -04001918 err = do_path_lookup(AT_FDCWD, name, flags | LOOKUP_ROOT, &nd);
1919 if (!err)
1920 *path = nd.path;
1921 return err;
Josef 'Jeff' Sipek16f18202007-07-19 01:48:18 -07001922}
1923
James Morris057f6c02007-04-26 00:12:05 -07001924/*
1925 * Restricted form of lookup. Doesn't follow links, single-component only,
1926 * needs parent already locked. Doesn't follow mounts.
1927 * SMP-safe.
1928 */
Adrian Bunka244e162006-03-31 02:32:11 -08001929static struct dentry *lookup_hash(struct nameidata *nd)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001930{
Al Viro72bd8662012-06-10 17:17:17 -04001931 return __lookup_hash(&nd->last, nd->path.dentry, nd->flags);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001932}
1933
Christoph Hellwigeead1912007-10-16 23:25:38 -07001934/**
Randy Dunlapa6b91912008-03-19 17:01:00 -07001935 * lookup_one_len - filesystem helper to lookup single pathname component
Christoph Hellwigeead1912007-10-16 23:25:38 -07001936 * @name: pathname component to lookup
1937 * @base: base directory to lookup from
1938 * @len: maximum length @len should be interpreted to
1939 *
Randy Dunlapa6b91912008-03-19 17:01:00 -07001940 * Note that this routine is purely a helper for filesystem usage and should
1941 * not be called by generic code. Also note that by using this function the
Christoph Hellwigeead1912007-10-16 23:25:38 -07001942 * nameidata argument is passed to the filesystem methods and a filesystem
1943 * using this helper needs to be prepared for that.
1944 */
James Morris057f6c02007-04-26 00:12:05 -07001945struct dentry *lookup_one_len(const char *name, struct dentry *base, int len)
1946{
James Morris057f6c02007-04-26 00:12:05 -07001947 struct qstr this;
Al Viro6a96ba52011-03-07 23:49:20 -05001948 unsigned int c;
Miklos Szeredicda309d2012-03-26 12:54:21 +02001949 int err;
James Morris057f6c02007-04-26 00:12:05 -07001950
David Woodhouse2f9092e2009-04-20 23:18:37 +01001951 WARN_ON_ONCE(!mutex_is_locked(&base->d_inode->i_mutex));
1952
Al Viro6a96ba52011-03-07 23:49:20 -05001953 this.name = name;
1954 this.len = len;
Linus Torvalds0145acc2012-03-02 14:32:59 -08001955 this.hash = full_name_hash(name, len);
Al Viro6a96ba52011-03-07 23:49:20 -05001956 if (!len)
1957 return ERR_PTR(-EACCES);
1958
Al Viro6a96ba52011-03-07 23:49:20 -05001959 while (len--) {
1960 c = *(const unsigned char *)name++;
1961 if (c == '/' || c == '\0')
1962 return ERR_PTR(-EACCES);
Al Viro6a96ba52011-03-07 23:49:20 -05001963 }
Al Viro5a202bc2011-03-08 14:17:44 -05001964 /*
1965 * See if the low-level filesystem might want
1966 * to use its own hash..
1967 */
1968 if (base->d_flags & DCACHE_OP_HASH) {
1969 int err = base->d_op->d_hash(base, base->d_inode, &this);
1970 if (err < 0)
1971 return ERR_PTR(err);
1972 }
Christoph Hellwigeead1912007-10-16 23:25:38 -07001973
Miklos Szeredicda309d2012-03-26 12:54:21 +02001974 err = inode_permission(base->d_inode, MAY_EXEC);
1975 if (err)
1976 return ERR_PTR(err);
1977
Al Viro72bd8662012-06-10 17:17:17 -04001978 return __lookup_hash(&this, base, 0);
James Morris057f6c02007-04-26 00:12:05 -07001979}
1980
Andy Whitcroft1fa1e7f2011-11-02 09:44:39 +01001981int user_path_at_empty(int dfd, const char __user *name, unsigned flags,
1982 struct path *path, int *empty)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001983{
Al Viro2d8f3032008-07-22 09:59:21 -04001984 struct nameidata nd;
Andy Whitcroft1fa1e7f2011-11-02 09:44:39 +01001985 char *tmp = getname_flags(name, flags, empty);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001986 int err = PTR_ERR(tmp);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001987 if (!IS_ERR(tmp)) {
Al Viro2d8f3032008-07-22 09:59:21 -04001988
1989 BUG_ON(flags & LOOKUP_PARENT);
1990
1991 err = do_path_lookup(dfd, tmp, flags, &nd);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001992 putname(tmp);
Al Viro2d8f3032008-07-22 09:59:21 -04001993 if (!err)
1994 *path = nd.path;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001995 }
1996 return err;
1997}
1998
Andy Whitcroft1fa1e7f2011-11-02 09:44:39 +01001999int user_path_at(int dfd, const char __user *name, unsigned flags,
2000 struct path *path)
2001{
Linus Torvaldsf7493e52012-03-22 16:10:40 -07002002 return user_path_at_empty(dfd, name, flags, path, NULL);
Andy Whitcroft1fa1e7f2011-11-02 09:44:39 +01002003}
2004
Al Viro2ad94ae2008-07-21 09:32:51 -04002005static int user_path_parent(int dfd, const char __user *path,
2006 struct nameidata *nd, char **name)
2007{
2008 char *s = getname(path);
2009 int error;
2010
2011 if (IS_ERR(s))
2012 return PTR_ERR(s);
2013
2014 error = do_path_lookup(dfd, s, LOOKUP_PARENT, nd);
2015 if (error)
2016 putname(s);
2017 else
2018 *name = s;
2019
2020 return error;
2021}
2022
Linus Torvalds1da177e2005-04-16 15:20:36 -07002023/*
2024 * It's inline, so penalty for filesystems that don't use sticky bit is
2025 * minimal.
2026 */
2027static inline int check_sticky(struct inode *dir, struct inode *inode)
2028{
Eric W. Biederman8e96e3b2012-03-03 21:17:15 -08002029 kuid_t fsuid = current_fsuid();
David Howellsda9592e2008-11-14 10:39:05 +11002030
Linus Torvalds1da177e2005-04-16 15:20:36 -07002031 if (!(dir->i_mode & S_ISVTX))
2032 return 0;
Eric W. Biederman8e96e3b2012-03-03 21:17:15 -08002033 if (uid_eq(inode->i_uid, fsuid))
Linus Torvalds1da177e2005-04-16 15:20:36 -07002034 return 0;
Eric W. Biederman8e96e3b2012-03-03 21:17:15 -08002035 if (uid_eq(dir->i_uid, fsuid))
Linus Torvalds1da177e2005-04-16 15:20:36 -07002036 return 0;
Eric W. Biederman1a48e2a2011-11-14 16:24:06 -08002037 return !inode_capable(inode, CAP_FOWNER);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002038}
2039
2040/*
2041 * Check whether we can remove a link victim from directory dir, check
2042 * whether the type of victim is right.
2043 * 1. We can't do it if dir is read-only (done in permission())
2044 * 2. We should have write and exec permissions on dir
2045 * 3. We can't remove anything from append-only dir
2046 * 4. We can't do anything with immutable dir (done in permission())
2047 * 5. If the sticky bit on dir is set we should either
2048 * a. be owner of dir, or
2049 * b. be owner of victim, or
2050 * c. have CAP_FOWNER capability
2051 * 6. If the victim is append-only or immutable we can't do antyhing with
2052 * links pointing to it.
2053 * 7. If we were asked to remove a directory and victim isn't one - ENOTDIR.
2054 * 8. If we were asked to remove a non-directory and victim isn't one - EISDIR.
2055 * 9. We can't remove a root or mountpoint.
2056 * 10. We don't allow removal of NFS sillyrenamed files; it's handled by
2057 * nfs_async_unlink().
2058 */
Arjan van de Ven858119e2006-01-14 13:20:43 -08002059static int may_delete(struct inode *dir,struct dentry *victim,int isdir)
Linus Torvalds1da177e2005-04-16 15:20:36 -07002060{
2061 int error;
2062
2063 if (!victim->d_inode)
2064 return -ENOENT;
2065
2066 BUG_ON(victim->d_parent->d_inode != dir);
Al Virocccc6bb2009-12-25 05:07:33 -05002067 audit_inode_child(victim, dir);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002068
Al Virof419a2e2008-07-22 00:07:17 -04002069 error = inode_permission(dir, MAY_WRITE | MAY_EXEC);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002070 if (error)
2071 return error;
2072 if (IS_APPEND(dir))
2073 return -EPERM;
2074 if (check_sticky(dir, victim->d_inode)||IS_APPEND(victim->d_inode)||
Hugh Dickinsf9454542008-11-19 15:36:38 -08002075 IS_IMMUTABLE(victim->d_inode) || IS_SWAPFILE(victim->d_inode))
Linus Torvalds1da177e2005-04-16 15:20:36 -07002076 return -EPERM;
2077 if (isdir) {
2078 if (!S_ISDIR(victim->d_inode->i_mode))
2079 return -ENOTDIR;
2080 if (IS_ROOT(victim))
2081 return -EBUSY;
2082 } else if (S_ISDIR(victim->d_inode->i_mode))
2083 return -EISDIR;
2084 if (IS_DEADDIR(dir))
2085 return -ENOENT;
2086 if (victim->d_flags & DCACHE_NFSFS_RENAMED)
2087 return -EBUSY;
2088 return 0;
2089}
2090
2091/* Check whether we can create an object with dentry child in directory
2092 * dir.
2093 * 1. We can't do it if child already exists (open has special treatment for
2094 * this case, but since we are inlined it's OK)
2095 * 2. We can't do it if dir is read-only (done in permission())
2096 * 3. We should have write and exec permissions on dir
2097 * 4. We can't do it if dir is immutable (done in permission())
2098 */
Miklos Szeredia95164d2008-07-30 15:08:48 +02002099static inline int may_create(struct inode *dir, struct dentry *child)
Linus Torvalds1da177e2005-04-16 15:20:36 -07002100{
2101 if (child->d_inode)
2102 return -EEXIST;
2103 if (IS_DEADDIR(dir))
2104 return -ENOENT;
Al Virof419a2e2008-07-22 00:07:17 -04002105 return inode_permission(dir, MAY_WRITE | MAY_EXEC);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002106}
2107
Linus Torvalds1da177e2005-04-16 15:20:36 -07002108/*
2109 * p1 and p2 should be directories on the same fs.
2110 */
2111struct dentry *lock_rename(struct dentry *p1, struct dentry *p2)
2112{
2113 struct dentry *p;
2114
2115 if (p1 == p2) {
Ingo Molnarf2eace22006-07-03 00:25:05 -07002116 mutex_lock_nested(&p1->d_inode->i_mutex, I_MUTEX_PARENT);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002117 return NULL;
2118 }
2119
Arjan van de Vena11f3a02006-03-23 03:00:33 -08002120 mutex_lock(&p1->d_inode->i_sb->s_vfs_rename_mutex);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002121
OGAWA Hirofumie2761a12008-10-16 07:50:28 +09002122 p = d_ancestor(p2, p1);
2123 if (p) {
2124 mutex_lock_nested(&p2->d_inode->i_mutex, I_MUTEX_PARENT);
2125 mutex_lock_nested(&p1->d_inode->i_mutex, I_MUTEX_CHILD);
2126 return p;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002127 }
2128
OGAWA Hirofumie2761a12008-10-16 07:50:28 +09002129 p = d_ancestor(p1, p2);
2130 if (p) {
2131 mutex_lock_nested(&p1->d_inode->i_mutex, I_MUTEX_PARENT);
2132 mutex_lock_nested(&p2->d_inode->i_mutex, I_MUTEX_CHILD);
2133 return p;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002134 }
2135
Ingo Molnarf2eace22006-07-03 00:25:05 -07002136 mutex_lock_nested(&p1->d_inode->i_mutex, I_MUTEX_PARENT);
2137 mutex_lock_nested(&p2->d_inode->i_mutex, I_MUTEX_CHILD);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002138 return NULL;
2139}
2140
2141void unlock_rename(struct dentry *p1, struct dentry *p2)
2142{
Jes Sorensen1b1dcc12006-01-09 15:59:24 -08002143 mutex_unlock(&p1->d_inode->i_mutex);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002144 if (p1 != p2) {
Jes Sorensen1b1dcc12006-01-09 15:59:24 -08002145 mutex_unlock(&p2->d_inode->i_mutex);
Arjan van de Vena11f3a02006-03-23 03:00:33 -08002146 mutex_unlock(&p1->d_inode->i_sb->s_vfs_rename_mutex);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002147 }
2148}
2149
Al Viro4acdaf22011-07-26 01:42:34 -04002150int vfs_create(struct inode *dir, struct dentry *dentry, umode_t mode,
Al Viro312b63f2012-06-10 18:09:36 -04002151 bool want_excl)
Linus Torvalds1da177e2005-04-16 15:20:36 -07002152{
Miklos Szeredia95164d2008-07-30 15:08:48 +02002153 int error = may_create(dir, dentry);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002154 if (error)
2155 return error;
2156
Al Viroacfa4382008-12-04 10:06:33 -05002157 if (!dir->i_op->create)
Linus Torvalds1da177e2005-04-16 15:20:36 -07002158 return -EACCES; /* shouldn't it be ENOSYS? */
2159 mode &= S_IALLUGO;
2160 mode |= S_IFREG;
2161 error = security_inode_create(dir, dentry, mode);
2162 if (error)
2163 return error;
Al Viro312b63f2012-06-10 18:09:36 -04002164 error = dir->i_op->create(dir, dentry, mode, want_excl);
Stephen Smalleya74574a2005-09-09 13:01:44 -07002165 if (!error)
Amy Griffisf38aa942005-11-03 15:57:06 +00002166 fsnotify_create(dir, dentry);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002167 return error;
2168}
2169
Al Viro73d049a2011-03-11 12:08:24 -05002170static int may_open(struct path *path, int acc_mode, int flag)
Linus Torvalds1da177e2005-04-16 15:20:36 -07002171{
Christoph Hellwig3fb64192008-10-24 09:58:10 +02002172 struct dentry *dentry = path->dentry;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002173 struct inode *inode = dentry->d_inode;
2174 int error;
2175
Al Virobcda7652011-03-13 16:42:14 -04002176 /* O_PATH? */
2177 if (!acc_mode)
2178 return 0;
2179
Linus Torvalds1da177e2005-04-16 15:20:36 -07002180 if (!inode)
2181 return -ENOENT;
2182
Christoph Hellwigc8fe8f32009-01-05 19:27:23 +01002183 switch (inode->i_mode & S_IFMT) {
2184 case S_IFLNK:
Linus Torvalds1da177e2005-04-16 15:20:36 -07002185 return -ELOOP;
Christoph Hellwigc8fe8f32009-01-05 19:27:23 +01002186 case S_IFDIR:
2187 if (acc_mode & MAY_WRITE)
2188 return -EISDIR;
2189 break;
2190 case S_IFBLK:
2191 case S_IFCHR:
Christoph Hellwig3fb64192008-10-24 09:58:10 +02002192 if (path->mnt->mnt_flags & MNT_NODEV)
Linus Torvalds1da177e2005-04-16 15:20:36 -07002193 return -EACCES;
Christoph Hellwigc8fe8f32009-01-05 19:27:23 +01002194 /*FALLTHRU*/
2195 case S_IFIFO:
2196 case S_IFSOCK:
Linus Torvalds1da177e2005-04-16 15:20:36 -07002197 flag &= ~O_TRUNC;
Christoph Hellwigc8fe8f32009-01-05 19:27:23 +01002198 break;
Dave Hansen4a3fd212008-02-15 14:37:48 -08002199 }
Dave Hansenb41572e2007-10-16 23:31:14 -07002200
Christoph Hellwig3fb64192008-10-24 09:58:10 +02002201 error = inode_permission(inode, acc_mode);
Dave Hansenb41572e2007-10-16 23:31:14 -07002202 if (error)
2203 return error;
Mimi Zohar6146f0d2009-02-04 09:06:57 -05002204
Linus Torvalds1da177e2005-04-16 15:20:36 -07002205 /*
2206 * An append-only file must be opened in append mode for writing.
2207 */
2208 if (IS_APPEND(inode)) {
Al Viro8737c932009-12-24 06:47:55 -05002209 if ((flag & O_ACCMODE) != O_RDONLY && !(flag & O_APPEND))
Al Viro7715b522009-12-16 03:54:00 -05002210 return -EPERM;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002211 if (flag & O_TRUNC)
Al Viro7715b522009-12-16 03:54:00 -05002212 return -EPERM;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002213 }
2214
2215 /* O_NOATIME can only be set by the owner or superuser */
Serge E. Hallyn2e149672011-03-23 16:43:26 -07002216 if (flag & O_NOATIME && !inode_owner_or_capable(inode))
Al Viro7715b522009-12-16 03:54:00 -05002217 return -EPERM;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002218
J. Bruce Fieldsf3c7691e2011-09-21 10:58:13 -04002219 return 0;
Al Viro7715b522009-12-16 03:54:00 -05002220}
Linus Torvalds1da177e2005-04-16 15:20:36 -07002221
Jeff Laytone1181ee2010-12-07 16:19:50 -05002222static int handle_truncate(struct file *filp)
Al Viro7715b522009-12-16 03:54:00 -05002223{
Jeff Laytone1181ee2010-12-07 16:19:50 -05002224 struct path *path = &filp->f_path;
Al Viro7715b522009-12-16 03:54:00 -05002225 struct inode *inode = path->dentry->d_inode;
2226 int error = get_write_access(inode);
2227 if (error)
2228 return error;
2229 /*
2230 * Refuse to truncate files with mandatory locks held on them.
2231 */
2232 error = locks_verify_locked(inode);
2233 if (!error)
Tetsuo Handaea0d3ab2010-06-02 13:24:43 +09002234 error = security_path_truncate(path);
Al Viro7715b522009-12-16 03:54:00 -05002235 if (!error) {
2236 error = do_truncate(path->dentry, 0,
2237 ATTR_MTIME|ATTR_CTIME|ATTR_OPEN,
Jeff Laytone1181ee2010-12-07 16:19:50 -05002238 filp);
Al Viro7715b522009-12-16 03:54:00 -05002239 }
2240 put_write_access(inode);
Mimi Zoharacd0c932009-09-04 13:08:46 -04002241 return error;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002242}
2243
Dave Hansend57999e2008-02-15 14:37:27 -08002244static inline int open_to_namei_flags(int flag)
2245{
Al Viro8a5e9292011-06-25 19:15:54 -04002246 if ((flag & O_ACCMODE) == 3)
2247 flag--;
Dave Hansend57999e2008-02-15 14:37:27 -08002248 return flag;
2249}
2250
Miklos Szeredid18e9002012-06-05 15:10:17 +02002251static int may_o_create(struct path *dir, struct dentry *dentry, umode_t mode)
2252{
2253 int error = security_path_mknod(dir, dentry, mode, 0);
2254 if (error)
2255 return error;
2256
2257 error = inode_permission(dir->dentry->d_inode, MAY_WRITE | MAY_EXEC);
2258 if (error)
2259 return error;
2260
2261 return security_inode_create(dir->dentry->d_inode, dentry, mode);
2262}
2263
David Howells1acf0af2012-06-14 16:13:46 +01002264/*
2265 * Attempt to atomically look up, create and open a file from a negative
2266 * dentry.
2267 *
2268 * Returns 0 if successful. The file will have been created and attached to
2269 * @file by the filesystem calling finish_open().
2270 *
2271 * Returns 1 if the file was looked up only or didn't need creating. The
2272 * caller will need to perform the open themselves. @path will have been
2273 * updated to point to the new dentry. This may be negative.
2274 *
2275 * Returns an error code otherwise.
2276 */
Al Viro2675a4e2012-06-22 12:41:10 +04002277static int atomic_open(struct nameidata *nd, struct dentry *dentry,
2278 struct path *path, struct file *file,
2279 const struct open_flags *op,
2280 bool *want_write, bool need_lookup,
2281 int *opened)
Miklos Szeredid18e9002012-06-05 15:10:17 +02002282{
2283 struct inode *dir = nd->path.dentry->d_inode;
2284 unsigned open_flag = open_to_namei_flags(op->open_flag);
2285 umode_t mode;
2286 int error;
2287 int acc_mode;
Miklos Szeredid18e9002012-06-05 15:10:17 +02002288 int create_error = 0;
2289 struct dentry *const DENTRY_NOT_SET = (void *) -1UL;
2290
2291 BUG_ON(dentry->d_inode);
2292
2293 /* Don't create child dentry for a dead directory. */
2294 if (unlikely(IS_DEADDIR(dir))) {
Al Viro2675a4e2012-06-22 12:41:10 +04002295 error = -ENOENT;
Miklos Szeredid18e9002012-06-05 15:10:17 +02002296 goto out;
2297 }
2298
2299 mode = op->mode & S_IALLUGO;
2300 if ((open_flag & O_CREAT) && !IS_POSIXACL(dir))
2301 mode &= ~current_umask();
2302
2303 if (open_flag & O_EXCL) {
2304 open_flag &= ~O_TRUNC;
Al Viro47237682012-06-10 05:01:45 -04002305 *opened |= FILE_CREATED;
Miklos Szeredid18e9002012-06-05 15:10:17 +02002306 }
2307
2308 /*
2309 * Checking write permission is tricky, bacuse we don't know if we are
2310 * going to actually need it: O_CREAT opens should work as long as the
2311 * file exists. But checking existence breaks atomicity. The trick is
2312 * to check access and if not granted clear O_CREAT from the flags.
2313 *
2314 * Another problem is returing the "right" error value (e.g. for an
2315 * O_EXCL open we want to return EEXIST not EROFS).
2316 */
2317 if ((open_flag & (O_CREAT | O_TRUNC)) ||
2318 (open_flag & O_ACCMODE) != O_RDONLY) {
2319 error = mnt_want_write(nd->path.mnt);
2320 if (!error) {
Miklos Szeredi77d660a2012-06-05 15:10:30 +02002321 *want_write = true;
Miklos Szeredid18e9002012-06-05 15:10:17 +02002322 } else if (!(open_flag & O_CREAT)) {
2323 /*
2324 * No O_CREATE -> atomicity not a requirement -> fall
2325 * back to lookup + open
2326 */
2327 goto no_open;
2328 } else if (open_flag & (O_EXCL | O_TRUNC)) {
2329 /* Fall back and fail with the right error */
2330 create_error = error;
2331 goto no_open;
2332 } else {
2333 /* No side effects, safe to clear O_CREAT */
2334 create_error = error;
2335 open_flag &= ~O_CREAT;
2336 }
2337 }
2338
2339 if (open_flag & O_CREAT) {
2340 error = may_o_create(&nd->path, dentry, op->mode);
2341 if (error) {
2342 create_error = error;
2343 if (open_flag & O_EXCL)
2344 goto no_open;
2345 open_flag &= ~O_CREAT;
2346 }
2347 }
2348
2349 if (nd->flags & LOOKUP_DIRECTORY)
2350 open_flag |= O_DIRECTORY;
2351
Al Viro30d90492012-06-22 12:40:19 +04002352 file->f_path.dentry = DENTRY_NOT_SET;
2353 file->f_path.mnt = nd->path.mnt;
2354 error = dir->i_op->atomic_open(dir, dentry, file, open_flag, mode,
Al Viro47237682012-06-10 05:01:45 -04002355 opened);
Al Virod9585272012-06-22 12:39:14 +04002356 if (error < 0) {
Al Virod9585272012-06-22 12:39:14 +04002357 if (create_error && error == -ENOENT)
2358 error = create_error;
Miklos Szeredid18e9002012-06-05 15:10:17 +02002359 goto out;
2360 }
2361
2362 acc_mode = op->acc_mode;
Al Viro47237682012-06-10 05:01:45 -04002363 if (*opened & FILE_CREATED) {
Miklos Szeredid18e9002012-06-05 15:10:17 +02002364 fsnotify_create(dir, dentry);
2365 acc_mode = MAY_OPEN;
2366 }
2367
Al Virod9585272012-06-22 12:39:14 +04002368 if (error) { /* returned 1, that is */
Al Viro30d90492012-06-22 12:40:19 +04002369 if (WARN_ON(file->f_path.dentry == DENTRY_NOT_SET)) {
Al Viro2675a4e2012-06-22 12:41:10 +04002370 error = -EIO;
Miklos Szeredid18e9002012-06-05 15:10:17 +02002371 goto out;
2372 }
Al Viro30d90492012-06-22 12:40:19 +04002373 if (file->f_path.dentry) {
Miklos Szeredid18e9002012-06-05 15:10:17 +02002374 dput(dentry);
Al Viro30d90492012-06-22 12:40:19 +04002375 dentry = file->f_path.dentry;
Miklos Szeredid18e9002012-06-05 15:10:17 +02002376 }
2377 goto looked_up;
2378 }
2379
2380 /*
2381 * We didn't have the inode before the open, so check open permission
2382 * here.
2383 */
Al Viro2675a4e2012-06-22 12:41:10 +04002384 error = may_open(&file->f_path, acc_mode, open_flag);
2385 if (error)
2386 fput(file);
Miklos Szeredid18e9002012-06-05 15:10:17 +02002387
2388out:
2389 dput(dentry);
Al Viro2675a4e2012-06-22 12:41:10 +04002390 return error;
Miklos Szeredid18e9002012-06-05 15:10:17 +02002391
Miklos Szeredid18e9002012-06-05 15:10:17 +02002392no_open:
2393 if (need_lookup) {
Al Viro72bd8662012-06-10 17:17:17 -04002394 dentry = lookup_real(dir, dentry, nd->flags);
Miklos Szeredid18e9002012-06-05 15:10:17 +02002395 if (IS_ERR(dentry))
Al Viro2675a4e2012-06-22 12:41:10 +04002396 return PTR_ERR(dentry);
Miklos Szeredid18e9002012-06-05 15:10:17 +02002397
2398 if (create_error) {
2399 int open_flag = op->open_flag;
2400
Al Viro2675a4e2012-06-22 12:41:10 +04002401 error = create_error;
Miklos Szeredid18e9002012-06-05 15:10:17 +02002402 if ((open_flag & O_EXCL)) {
2403 if (!dentry->d_inode)
2404 goto out;
2405 } else if (!dentry->d_inode) {
2406 goto out;
2407 } else if ((open_flag & O_TRUNC) &&
2408 S_ISREG(dentry->d_inode->i_mode)) {
2409 goto out;
2410 }
2411 /* will fail later, go on to get the right error */
2412 }
2413 }
2414looked_up:
2415 path->dentry = dentry;
2416 path->mnt = nd->path.mnt;
Al Viro2675a4e2012-06-22 12:41:10 +04002417 return 1;
Miklos Szeredid18e9002012-06-05 15:10:17 +02002418}
2419
Nick Piggin31e6b012011-01-07 17:49:52 +11002420/*
David Howells1acf0af2012-06-14 16:13:46 +01002421 * Look up and maybe create and open the last component.
Miklos Szeredid58ffd32012-06-05 15:10:15 +02002422 *
2423 * Must be called with i_mutex held on parent.
2424 *
David Howells1acf0af2012-06-14 16:13:46 +01002425 * Returns 0 if the file was successfully atomically created (if necessary) and
2426 * opened. In this case the file will be returned attached to @file.
2427 *
2428 * Returns 1 if the file was not completely opened at this time, though lookups
2429 * and creations will have been performed and the dentry returned in @path will
2430 * be positive upon return if O_CREAT was specified. If O_CREAT wasn't
2431 * specified then a negative dentry may be returned.
2432 *
2433 * An error code is returned otherwise.
2434 *
2435 * FILE_CREATE will be set in @*opened if the dentry was created and will be
2436 * cleared otherwise prior to returning.
Miklos Szeredid58ffd32012-06-05 15:10:15 +02002437 */
Al Viro2675a4e2012-06-22 12:41:10 +04002438static int lookup_open(struct nameidata *nd, struct path *path,
2439 struct file *file,
2440 const struct open_flags *op,
2441 bool *want_write, int *opened)
Miklos Szeredid58ffd32012-06-05 15:10:15 +02002442{
2443 struct dentry *dir = nd->path.dentry;
Miklos Szeredi54ef4872012-06-05 15:10:16 +02002444 struct inode *dir_inode = dir->d_inode;
Miklos Szeredid58ffd32012-06-05 15:10:15 +02002445 struct dentry *dentry;
2446 int error;
Miklos Szeredi54ef4872012-06-05 15:10:16 +02002447 bool need_lookup;
Miklos Szeredid58ffd32012-06-05 15:10:15 +02002448
Al Viro47237682012-06-10 05:01:45 -04002449 *opened &= ~FILE_CREATED;
Al Viro201f9562012-06-22 12:42:10 +04002450 dentry = lookup_dcache(&nd->last, dir, nd->flags, &need_lookup);
Miklos Szeredid58ffd32012-06-05 15:10:15 +02002451 if (IS_ERR(dentry))
Al Viro2675a4e2012-06-22 12:41:10 +04002452 return PTR_ERR(dentry);
Miklos Szeredid58ffd32012-06-05 15:10:15 +02002453
Miklos Szeredid18e9002012-06-05 15:10:17 +02002454 /* Cached positive dentry: will open in f_op->open */
2455 if (!need_lookup && dentry->d_inode)
2456 goto out_no_open;
2457
2458 if ((nd->flags & LOOKUP_OPEN) && dir_inode->i_op->atomic_open) {
Al Viro30d90492012-06-22 12:40:19 +04002459 return atomic_open(nd, dentry, path, file, op, want_write,
Al Viro47237682012-06-10 05:01:45 -04002460 need_lookup, opened);
Miklos Szeredid18e9002012-06-05 15:10:17 +02002461 }
2462
Miklos Szeredi54ef4872012-06-05 15:10:16 +02002463 if (need_lookup) {
2464 BUG_ON(dentry->d_inode);
2465
Al Viro72bd8662012-06-10 17:17:17 -04002466 dentry = lookup_real(dir_inode, dentry, nd->flags);
Miklos Szeredi54ef4872012-06-05 15:10:16 +02002467 if (IS_ERR(dentry))
Al Viro2675a4e2012-06-22 12:41:10 +04002468 return PTR_ERR(dentry);
Miklos Szeredi54ef4872012-06-05 15:10:16 +02002469 }
2470
Miklos Szeredid58ffd32012-06-05 15:10:15 +02002471 /* Negative dentry, just create the file */
2472 if (!dentry->d_inode && (op->open_flag & O_CREAT)) {
2473 umode_t mode = op->mode;
2474 if (!IS_POSIXACL(dir->d_inode))
2475 mode &= ~current_umask();
2476 /*
2477 * This write is needed to ensure that a
2478 * rw->ro transition does not occur between
2479 * the time when the file is created and when
2480 * a permanent write count is taken through
Miklos Szeredi015c3bb2012-06-05 15:10:27 +02002481 * the 'struct file' in finish_open().
Miklos Szeredid58ffd32012-06-05 15:10:15 +02002482 */
2483 error = mnt_want_write(nd->path.mnt);
2484 if (error)
2485 goto out_dput;
Miklos Szeredi77d660a2012-06-05 15:10:30 +02002486 *want_write = true;
Al Viro47237682012-06-10 05:01:45 -04002487 *opened |= FILE_CREATED;
Miklos Szeredid58ffd32012-06-05 15:10:15 +02002488 error = security_path_mknod(&nd->path, dentry, mode, 0);
2489 if (error)
2490 goto out_dput;
Al Viro312b63f2012-06-10 18:09:36 -04002491 error = vfs_create(dir->d_inode, dentry, mode,
2492 nd->flags & LOOKUP_EXCL);
Miklos Szeredid58ffd32012-06-05 15:10:15 +02002493 if (error)
2494 goto out_dput;
2495 }
Miklos Szeredid18e9002012-06-05 15:10:17 +02002496out_no_open:
Miklos Szeredid58ffd32012-06-05 15:10:15 +02002497 path->dentry = dentry;
2498 path->mnt = nd->path.mnt;
Al Viro2675a4e2012-06-22 12:41:10 +04002499 return 1;
Miklos Szeredid58ffd32012-06-05 15:10:15 +02002500
2501out_dput:
2502 dput(dentry);
Al Viro2675a4e2012-06-22 12:41:10 +04002503 return error;
Miklos Szeredid58ffd32012-06-05 15:10:15 +02002504}
2505
2506/*
Al Virofe2d35f2011-03-05 22:58:25 -05002507 * Handle the last step of open()
Nick Piggin31e6b012011-01-07 17:49:52 +11002508 */
Al Viro2675a4e2012-06-22 12:41:10 +04002509static int do_last(struct nameidata *nd, struct path *path,
2510 struct file *file, const struct open_flags *op,
2511 int *opened, const char *pathname)
Al Virofb1cc552009-12-24 01:58:28 -05002512{
Al Viroa1e28032009-12-24 02:12:06 -05002513 struct dentry *dir = nd->path.dentry;
Al Viroca344a892011-03-09 00:36:45 -05002514 int open_flag = op->open_flag;
Miklos Szeredi77d660a2012-06-05 15:10:30 +02002515 bool will_truncate = (open_flag & O_TRUNC) != 0;
2516 bool want_write = false;
Al Virobcda7652011-03-13 16:42:14 -04002517 int acc_mode = op->acc_mode;
Miklos Szeredia1eb3312012-05-21 17:30:07 +02002518 struct inode *inode;
Miklos Szeredi77d660a2012-06-05 15:10:30 +02002519 bool symlink_ok = false;
Miklos Szeredi16b1c1c2012-05-21 17:30:19 +02002520 struct path save_parent = { .dentry = NULL, .mnt = NULL };
2521 bool retried = false;
Al Viro16c2cd72011-02-22 15:50:10 -05002522 int error;
Al Virofb1cc552009-12-24 01:58:28 -05002523
Al Viroc3e380b2011-02-23 13:39:45 -05002524 nd->flags &= ~LOOKUP_PARENT;
2525 nd->flags |= op->intent;
2526
Al Viro1f36f772009-12-26 10:56:19 -05002527 switch (nd->last_type) {
2528 case LAST_DOTDOT:
Neil Brown176306f2010-05-24 16:57:56 +10002529 case LAST_DOT:
Al Virofe2d35f2011-03-05 22:58:25 -05002530 error = handle_dots(nd, nd->last_type);
2531 if (error)
Al Viro2675a4e2012-06-22 12:41:10 +04002532 return error;
Al Viro1f36f772009-12-26 10:56:19 -05002533 /* fallthrough */
Al Viro1f36f772009-12-26 10:56:19 -05002534 case LAST_ROOT:
Al Viro9f1fafe2011-03-25 11:00:12 -04002535 error = complete_walk(nd);
Al Viro16c2cd72011-02-22 15:50:10 -05002536 if (error)
Al Viro2675a4e2012-06-22 12:41:10 +04002537 return error;
Al Virofe2d35f2011-03-05 22:58:25 -05002538 audit_inode(pathname, nd->path.dentry);
Al Viroca344a892011-03-09 00:36:45 -05002539 if (open_flag & O_CREAT) {
Al Virofe2d35f2011-03-05 22:58:25 -05002540 error = -EISDIR;
Al Viro2675a4e2012-06-22 12:41:10 +04002541 goto out;
Al Virofe2d35f2011-03-05 22:58:25 -05002542 }
Miklos Szeredie83db162012-06-05 15:10:29 +02002543 goto finish_open;
Al Viro1f36f772009-12-26 10:56:19 -05002544 case LAST_BIND:
Al Viro9f1fafe2011-03-25 11:00:12 -04002545 error = complete_walk(nd);
Al Viro16c2cd72011-02-22 15:50:10 -05002546 if (error)
Al Viro2675a4e2012-06-22 12:41:10 +04002547 return error;
Al Viro1f36f772009-12-26 10:56:19 -05002548 audit_inode(pathname, dir);
Miklos Szeredie83db162012-06-05 15:10:29 +02002549 goto finish_open;
Al Viro1f36f772009-12-26 10:56:19 -05002550 }
Al Viro67ee3ad2009-12-26 07:01:01 -05002551
Al Viroca344a892011-03-09 00:36:45 -05002552 if (!(open_flag & O_CREAT)) {
Al Virofe2d35f2011-03-05 22:58:25 -05002553 if (nd->last.name[nd->last.len])
2554 nd->flags |= LOOKUP_FOLLOW | LOOKUP_DIRECTORY;
Al Virobcda7652011-03-13 16:42:14 -04002555 if (open_flag & O_PATH && !(nd->flags & LOOKUP_FOLLOW))
Miklos Szeredi77d660a2012-06-05 15:10:30 +02002556 symlink_ok = true;
Al Virofe2d35f2011-03-05 22:58:25 -05002557 /* we _can_ be in RCU mode here */
Miklos Szeredia1eb3312012-05-21 17:30:07 +02002558 error = lookup_fast(nd, &nd->last, path, &inode);
Miklos Szeredi71574862012-06-05 15:10:14 +02002559 if (likely(!error))
2560 goto finish_lookup;
Miklos Szeredia1eb3312012-05-21 17:30:07 +02002561
Miklos Szeredi71574862012-06-05 15:10:14 +02002562 if (error < 0)
Al Viro2675a4e2012-06-22 12:41:10 +04002563 goto out;
Miklos Szeredi37d7fff2012-06-05 15:10:12 +02002564
Miklos Szeredi71574862012-06-05 15:10:14 +02002565 BUG_ON(nd->inode != dir->d_inode);
Miklos Szeredib6183df2012-06-05 15:10:13 +02002566 } else {
2567 /* create side of things */
2568 /*
2569 * This will *only* deal with leaving RCU mode - LOOKUP_JUMPED
2570 * has been cleared when we got to the last component we are
2571 * about to look up
2572 */
2573 error = complete_walk(nd);
2574 if (error)
Al Viro2675a4e2012-06-22 12:41:10 +04002575 return error;
Miklos Szeredib6183df2012-06-05 15:10:13 +02002576
2577 audit_inode(pathname, dir);
2578 error = -EISDIR;
2579 /* trailing slashes? */
2580 if (nd->last.name[nd->last.len])
Al Viro2675a4e2012-06-22 12:41:10 +04002581 goto out;
Al Virofe2d35f2011-03-05 22:58:25 -05002582 }
2583
Miklos Szeredi16b1c1c2012-05-21 17:30:19 +02002584retry_lookup:
Al Viroa1e28032009-12-24 02:12:06 -05002585 mutex_lock(&dir->d_inode->i_mutex);
Al Viro2675a4e2012-06-22 12:41:10 +04002586 error = lookup_open(nd, path, file, op, &want_write, opened);
Miklos Szeredid58ffd32012-06-05 15:10:15 +02002587 mutex_unlock(&dir->d_inode->i_mutex);
Al Viroa1e28032009-12-24 02:12:06 -05002588
Al Viro2675a4e2012-06-22 12:41:10 +04002589 if (error <= 0) {
2590 if (error)
Miklos Szeredid18e9002012-06-05 15:10:17 +02002591 goto out;
2592
Al Viro47237682012-06-10 05:01:45 -04002593 if ((*opened & FILE_CREATED) ||
Al Viro2675a4e2012-06-22 12:41:10 +04002594 !S_ISREG(file->f_path.dentry->d_inode->i_mode))
Miklos Szeredi77d660a2012-06-05 15:10:30 +02002595 will_truncate = false;
Miklos Szeredid18e9002012-06-05 15:10:17 +02002596
Al Viro2675a4e2012-06-22 12:41:10 +04002597 audit_inode(pathname, file->f_path.dentry);
Miklos Szeredid18e9002012-06-05 15:10:17 +02002598 goto opened;
2599 }
Al Virofb1cc552009-12-24 01:58:28 -05002600
Al Viro47237682012-06-10 05:01:45 -04002601 if (*opened & FILE_CREATED) {
Al Viro9b44f1b2011-03-09 00:17:27 -05002602 /* Don't check for write permission, don't truncate */
Al Viroca344a892011-03-09 00:36:45 -05002603 open_flag &= ~O_TRUNC;
Miklos Szeredi77d660a2012-06-05 15:10:30 +02002604 will_truncate = false;
Al Virobcda7652011-03-13 16:42:14 -04002605 acc_mode = MAY_OPEN;
Miklos Szeredid58ffd32012-06-05 15:10:15 +02002606 path_to_nameidata(path, nd);
Miklos Szeredie83db162012-06-05 15:10:29 +02002607 goto finish_open_created;
Al Virofb1cc552009-12-24 01:58:28 -05002608 }
2609
2610 /*
2611 * It already exists.
2612 */
Al Virofb1cc552009-12-24 01:58:28 -05002613 audit_inode(pathname, path->dentry);
2614
Miklos Szeredid18e9002012-06-05 15:10:17 +02002615 /*
2616 * If atomic_open() acquired write access it is dropped now due to
2617 * possible mount and symlink following (this might be optimized away if
2618 * necessary...)
2619 */
2620 if (want_write) {
2621 mnt_drop_write(nd->path.mnt);
Miklos Szeredi77d660a2012-06-05 15:10:30 +02002622 want_write = false;
Miklos Szeredid18e9002012-06-05 15:10:17 +02002623 }
2624
Al Virofb1cc552009-12-24 01:58:28 -05002625 error = -EEXIST;
Al Viroca344a892011-03-09 00:36:45 -05002626 if (open_flag & O_EXCL)
Al Virofb1cc552009-12-24 01:58:28 -05002627 goto exit_dput;
2628
David Howells9875cf82011-01-14 18:45:21 +00002629 error = follow_managed(path, nd->flags);
2630 if (error < 0)
2631 goto exit_dput;
Al Virofb1cc552009-12-24 01:58:28 -05002632
Al Viroa3fbbde2011-11-07 21:21:26 +00002633 if (error)
2634 nd->flags |= LOOKUP_JUMPED;
2635
Miklos Szeredidecf3402012-05-21 17:30:08 +02002636 BUG_ON(nd->flags & LOOKUP_RCU);
2637 inode = path->dentry->d_inode;
Miklos Szeredi5f5daac2012-05-21 17:30:14 +02002638finish_lookup:
2639 /* we _can_ be in RCU mode here */
Al Virofb1cc552009-12-24 01:58:28 -05002640 error = -ENOENT;
Miklos Szeredi54c33e72012-05-21 17:30:10 +02002641 if (!inode) {
2642 path_to_nameidata(path, nd);
Al Viro2675a4e2012-06-22 12:41:10 +04002643 goto out;
Miklos Szeredi54c33e72012-05-21 17:30:10 +02002644 }
Al Viro9e67f362009-12-26 07:04:50 -05002645
Miklos Szeredid45ea862012-05-21 17:30:09 +02002646 if (should_follow_link(inode, !symlink_ok)) {
2647 if (nd->flags & LOOKUP_RCU) {
2648 if (unlikely(unlazy_walk(nd, path->dentry))) {
2649 error = -ECHILD;
Al Viro2675a4e2012-06-22 12:41:10 +04002650 goto out;
Miklos Szeredid45ea862012-05-21 17:30:09 +02002651 }
2652 }
2653 BUG_ON(inode != path->dentry->d_inode);
Al Viro2675a4e2012-06-22 12:41:10 +04002654 return 1;
Miklos Szeredid45ea862012-05-21 17:30:09 +02002655 }
Al Virofb1cc552009-12-24 01:58:28 -05002656
Miklos Szeredi16b1c1c2012-05-21 17:30:19 +02002657 if ((nd->flags & LOOKUP_RCU) || nd->path.mnt != path->mnt) {
2658 path_to_nameidata(path, nd);
2659 } else {
2660 save_parent.dentry = nd->path.dentry;
2661 save_parent.mnt = mntget(path->mnt);
2662 nd->path.dentry = path->dentry;
2663
2664 }
Miklos Szeredidecf3402012-05-21 17:30:08 +02002665 nd->inode = inode;
Al Viroa3fbbde2011-11-07 21:21:26 +00002666 /* Why this, you ask? _Now_ we might have grown LOOKUP_JUMPED... */
2667 error = complete_walk(nd);
Miklos Szeredi16b1c1c2012-05-21 17:30:19 +02002668 if (error) {
2669 path_put(&save_parent);
Al Viro2675a4e2012-06-22 12:41:10 +04002670 return error;
Miklos Szeredi16b1c1c2012-05-21 17:30:19 +02002671 }
Al Virofb1cc552009-12-24 01:58:28 -05002672 error = -EISDIR;
Miklos Szeredi050ac842012-05-21 17:30:12 +02002673 if ((open_flag & O_CREAT) && S_ISDIR(nd->inode->i_mode))
Al Viro2675a4e2012-06-22 12:41:10 +04002674 goto out;
Miklos Szerediaf2f5542012-05-21 17:30:11 +02002675 error = -ENOTDIR;
2676 if ((nd->flags & LOOKUP_DIRECTORY) && !nd->inode->i_op->lookup)
Al Viro2675a4e2012-06-22 12:41:10 +04002677 goto out;
Miklos Szeredid7fdd7f2012-05-21 17:30:13 +02002678 audit_inode(pathname, nd->path.dentry);
Miklos Szeredie83db162012-06-05 15:10:29 +02002679finish_open:
Al Viro6c0d46c2011-03-09 00:59:59 -05002680 if (!S_ISREG(nd->inode->i_mode))
Miklos Szeredi77d660a2012-06-05 15:10:30 +02002681 will_truncate = false;
Al Viro6c0d46c2011-03-09 00:59:59 -05002682
Al Viro0f9d1a12011-03-09 00:13:14 -05002683 if (will_truncate) {
2684 error = mnt_want_write(nd->path.mnt);
2685 if (error)
Al Viro2675a4e2012-06-22 12:41:10 +04002686 goto out;
Miklos Szeredi77d660a2012-06-05 15:10:30 +02002687 want_write = true;
Al Viro0f9d1a12011-03-09 00:13:14 -05002688 }
Miklos Szeredie83db162012-06-05 15:10:29 +02002689finish_open_created:
Al Virobcda7652011-03-13 16:42:14 -04002690 error = may_open(&nd->path, acc_mode, open_flag);
Al Viroca344a892011-03-09 00:36:45 -05002691 if (error)
Al Viro2675a4e2012-06-22 12:41:10 +04002692 goto out;
Al Viro30d90492012-06-22 12:40:19 +04002693 file->f_path.mnt = nd->path.mnt;
2694 error = finish_open(file, nd->path.dentry, NULL, opened);
2695 if (error) {
Al Viro30d90492012-06-22 12:40:19 +04002696 if (error == -EOPENSTALE)
Miklos Szeredif60dc3d2012-06-05 15:10:31 +02002697 goto stale_open;
Miklos Szeredi015c3bb2012-06-05 15:10:27 +02002698 goto out;
Miklos Szeredif60dc3d2012-06-05 15:10:31 +02002699 }
Miklos Szeredia8277b92012-06-05 15:10:32 +02002700opened:
Al Viro2675a4e2012-06-22 12:41:10 +04002701 error = open_check_o_direct(file);
Miklos Szeredi015c3bb2012-06-05 15:10:27 +02002702 if (error)
2703 goto exit_fput;
Al Viro2675a4e2012-06-22 12:41:10 +04002704 error = ima_file_check(file, op->acc_mode);
Miklos Szerediaa4caad2012-06-05 15:10:28 +02002705 if (error)
2706 goto exit_fput;
2707
2708 if (will_truncate) {
Al Viro2675a4e2012-06-22 12:41:10 +04002709 error = handle_truncate(file);
Miklos Szerediaa4caad2012-06-05 15:10:28 +02002710 if (error)
2711 goto exit_fput;
Al Viro0f9d1a12011-03-09 00:13:14 -05002712 }
Al Viroca344a892011-03-09 00:36:45 -05002713out:
2714 if (want_write)
Al Viro0f9d1a12011-03-09 00:13:14 -05002715 mnt_drop_write(nd->path.mnt);
Miklos Szeredi16b1c1c2012-05-21 17:30:19 +02002716 path_put(&save_parent);
Miklos Szeredie276ae62012-05-21 17:30:06 +02002717 terminate_walk(nd);
Al Viro2675a4e2012-06-22 12:41:10 +04002718 return error;
Al Virofb1cc552009-12-24 01:58:28 -05002719
Al Virofb1cc552009-12-24 01:58:28 -05002720exit_dput:
2721 path_put_conditional(path, nd);
Al Viroca344a892011-03-09 00:36:45 -05002722 goto out;
Miklos Szeredi015c3bb2012-06-05 15:10:27 +02002723exit_fput:
Al Viro2675a4e2012-06-22 12:41:10 +04002724 fput(file);
2725 goto out;
Miklos Szeredi015c3bb2012-06-05 15:10:27 +02002726
Miklos Szeredif60dc3d2012-06-05 15:10:31 +02002727stale_open:
2728 /* If no saved parent or already retried then can't retry */
2729 if (!save_parent.dentry || retried)
2730 goto out;
2731
2732 BUG_ON(save_parent.dentry != dir);
2733 path_put(&nd->path);
2734 nd->path = save_parent;
2735 nd->inode = dir->d_inode;
2736 save_parent.mnt = NULL;
2737 save_parent.dentry = NULL;
2738 if (want_write) {
2739 mnt_drop_write(nd->path.mnt);
2740 want_write = false;
2741 }
2742 retried = true;
2743 goto retry_lookup;
Al Virofb1cc552009-12-24 01:58:28 -05002744}
2745
Al Viro13aab422011-02-23 17:54:08 -05002746static struct file *path_openat(int dfd, const char *pathname,
Al Viro73d049a2011-03-11 12:08:24 -05002747 struct nameidata *nd, const struct open_flags *op, int flags)
Linus Torvalds1da177e2005-04-16 15:20:36 -07002748{
Al Virofe2d35f2011-03-05 22:58:25 -05002749 struct file *base = NULL;
Al Viro30d90492012-06-22 12:40:19 +04002750 struct file *file;
Al Viro9850c052010-01-13 15:01:15 -05002751 struct path path;
Al Viro47237682012-06-10 05:01:45 -04002752 int opened = 0;
Al Viro13aab422011-02-23 17:54:08 -05002753 int error;
Nick Piggin31e6b012011-01-07 17:49:52 +11002754
Al Viro30d90492012-06-22 12:40:19 +04002755 file = get_empty_filp();
2756 if (!file)
Nick Piggin31e6b012011-01-07 17:49:52 +11002757 return ERR_PTR(-ENFILE);
2758
Al Viro30d90492012-06-22 12:40:19 +04002759 file->f_flags = op->open_flag;
Nick Piggin31e6b012011-01-07 17:49:52 +11002760
Al Viro73d049a2011-03-11 12:08:24 -05002761 error = path_init(dfd, pathname, flags | LOOKUP_PARENT, nd, &base);
Nick Piggin31e6b012011-01-07 17:49:52 +11002762 if (unlikely(error))
Al Viro2675a4e2012-06-22 12:41:10 +04002763 goto out;
Nick Piggin31e6b012011-01-07 17:49:52 +11002764
Al Virofe2d35f2011-03-05 22:58:25 -05002765 current->total_link_count = 0;
Al Viro73d049a2011-03-11 12:08:24 -05002766 error = link_path_walk(pathname, nd);
Nick Piggin31e6b012011-01-07 17:49:52 +11002767 if (unlikely(error))
Al Viro2675a4e2012-06-22 12:41:10 +04002768 goto out;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002769
Al Viro2675a4e2012-06-22 12:41:10 +04002770 error = do_last(nd, &path, file, op, &opened, pathname);
2771 while (unlikely(error > 0)) { /* trailing symlink */
Nick Piggin7b9337a2011-01-14 08:42:43 +00002772 struct path link = path;
Al Virodef4af32009-12-26 08:37:05 -05002773 void *cookie;
Al Viro574197e2011-03-14 22:20:34 -04002774 if (!(nd->flags & LOOKUP_FOLLOW)) {
Al Viro73d049a2011-03-11 12:08:24 -05002775 path_put_conditional(&path, nd);
2776 path_put(&nd->path);
Al Viro2675a4e2012-06-22 12:41:10 +04002777 error = -ELOOP;
Al Viro40b39132011-03-09 16:22:18 -05002778 break;
2779 }
Al Viro73d049a2011-03-11 12:08:24 -05002780 nd->flags |= LOOKUP_PARENT;
2781 nd->flags &= ~(LOOKUP_OPEN|LOOKUP_CREATE|LOOKUP_EXCL);
Al Viro574197e2011-03-14 22:20:34 -04002782 error = follow_link(&link, nd, &cookie);
Al Viroc3e380b2011-02-23 13:39:45 -05002783 if (unlikely(error))
Al Viro2675a4e2012-06-22 12:41:10 +04002784 break;
2785 error = do_last(nd, &path, file, op, &opened, pathname);
Al Viro574197e2011-03-14 22:20:34 -04002786 put_link(nd, &link, cookie);
Al Viro806b6812009-12-26 07:16:40 -05002787 }
Al Viro10fa8e62009-12-26 07:09:49 -05002788out:
Al Viro73d049a2011-03-11 12:08:24 -05002789 if (nd->root.mnt && !(nd->flags & LOOKUP_ROOT))
2790 path_put(&nd->root);
Al Virofe2d35f2011-03-05 22:58:25 -05002791 if (base)
2792 fput(base);
Al Viro2675a4e2012-06-22 12:41:10 +04002793 if (!(opened & FILE_OPENED)) {
2794 BUG_ON(!error);
Al Viro30d90492012-06-22 12:40:19 +04002795 put_filp(file);
Miklos Szeredi015c3bb2012-06-05 15:10:27 +02002796 }
Al Viro2675a4e2012-06-22 12:41:10 +04002797 if (unlikely(error)) {
2798 if (error == -EOPENSTALE) {
2799 if (flags & LOOKUP_RCU)
2800 error = -ECHILD;
2801 else
2802 error = -ESTALE;
2803 }
2804 file = ERR_PTR(error);
2805 }
2806 return file;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002807}
2808
Al Viro13aab422011-02-23 17:54:08 -05002809struct file *do_filp_open(int dfd, const char *pathname,
2810 const struct open_flags *op, int flags)
2811{
Al Viro73d049a2011-03-11 12:08:24 -05002812 struct nameidata nd;
Al Viro13aab422011-02-23 17:54:08 -05002813 struct file *filp;
2814
Al Viro73d049a2011-03-11 12:08:24 -05002815 filp = path_openat(dfd, pathname, &nd, op, flags | LOOKUP_RCU);
Al Viro13aab422011-02-23 17:54:08 -05002816 if (unlikely(filp == ERR_PTR(-ECHILD)))
Al Viro73d049a2011-03-11 12:08:24 -05002817 filp = path_openat(dfd, pathname, &nd, op, flags);
Al Viro13aab422011-02-23 17:54:08 -05002818 if (unlikely(filp == ERR_PTR(-ESTALE)))
Al Viro73d049a2011-03-11 12:08:24 -05002819 filp = path_openat(dfd, pathname, &nd, op, flags | LOOKUP_REVAL);
Al Viro13aab422011-02-23 17:54:08 -05002820 return filp;
2821}
2822
Al Viro73d049a2011-03-11 12:08:24 -05002823struct file *do_file_open_root(struct dentry *dentry, struct vfsmount *mnt,
2824 const char *name, const struct open_flags *op, int flags)
2825{
2826 struct nameidata nd;
2827 struct file *file;
2828
2829 nd.root.mnt = mnt;
2830 nd.root.dentry = dentry;
2831
2832 flags |= LOOKUP_ROOT;
2833
Al Virobcda7652011-03-13 16:42:14 -04002834 if (dentry->d_inode->i_op->follow_link && op->intent & LOOKUP_OPEN)
Al Viro73d049a2011-03-11 12:08:24 -05002835 return ERR_PTR(-ELOOP);
2836
2837 file = path_openat(-1, name, &nd, op, flags | LOOKUP_RCU);
2838 if (unlikely(file == ERR_PTR(-ECHILD)))
2839 file = path_openat(-1, name, &nd, op, flags);
2840 if (unlikely(file == ERR_PTR(-ESTALE)))
2841 file = path_openat(-1, name, &nd, op, flags | LOOKUP_REVAL);
2842 return file;
2843}
2844
Al Viroed75e952011-06-27 16:53:43 -04002845struct dentry *kern_path_create(int dfd, const char *pathname, struct path *path, int is_dir)
Linus Torvalds1da177e2005-04-16 15:20:36 -07002846{
Christoph Hellwigc663e5d2005-06-23 00:09:49 -07002847 struct dentry *dentry = ERR_PTR(-EEXIST);
Al Viroed75e952011-06-27 16:53:43 -04002848 struct nameidata nd;
2849 int error = do_path_lookup(dfd, pathname, LOOKUP_PARENT, &nd);
2850 if (error)
2851 return ERR_PTR(error);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002852
Christoph Hellwigc663e5d2005-06-23 00:09:49 -07002853 /*
2854 * Yucky last component or no last component at all?
2855 * (foo/., foo/.., /////)
2856 */
Al Viroed75e952011-06-27 16:53:43 -04002857 if (nd.last_type != LAST_NORM)
2858 goto out;
2859 nd.flags &= ~LOOKUP_PARENT;
2860 nd.flags |= LOOKUP_CREATE | LOOKUP_EXCL;
Christoph Hellwigc663e5d2005-06-23 00:09:49 -07002861
2862 /*
2863 * Do the final lookup.
2864 */
Al Viroed75e952011-06-27 16:53:43 -04002865 mutex_lock_nested(&nd.path.dentry->d_inode->i_mutex, I_MUTEX_PARENT);
2866 dentry = lookup_hash(&nd);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002867 if (IS_ERR(dentry))
2868 goto fail;
Christoph Hellwigc663e5d2005-06-23 00:09:49 -07002869
Al Viroe9baf6e2008-05-15 04:49:12 -04002870 if (dentry->d_inode)
2871 goto eexist;
Christoph Hellwigc663e5d2005-06-23 00:09:49 -07002872 /*
2873 * Special case - lookup gave negative, but... we had foo/bar/
2874 * From the vfs_mknod() POV we just have a negative dentry -
2875 * all is fine. Let's be bastards - you had / on the end, you've
2876 * been asking for (non-existent) directory. -ENOENT for you.
2877 */
Al Viroed75e952011-06-27 16:53:43 -04002878 if (unlikely(!is_dir && nd.last.name[nd.last.len])) {
Al Viroe9baf6e2008-05-15 04:49:12 -04002879 dput(dentry);
2880 dentry = ERR_PTR(-ENOENT);
Al Viroed75e952011-06-27 16:53:43 -04002881 goto fail;
Al Viroe9baf6e2008-05-15 04:49:12 -04002882 }
Al Viroed75e952011-06-27 16:53:43 -04002883 *path = nd.path;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002884 return dentry;
Al Viroe9baf6e2008-05-15 04:49:12 -04002885eexist:
Linus Torvalds1da177e2005-04-16 15:20:36 -07002886 dput(dentry);
Al Viroe9baf6e2008-05-15 04:49:12 -04002887 dentry = ERR_PTR(-EEXIST);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002888fail:
Al Viroed75e952011-06-27 16:53:43 -04002889 mutex_unlock(&nd.path.dentry->d_inode->i_mutex);
2890out:
2891 path_put(&nd.path);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002892 return dentry;
2893}
Al Virodae6ad82011-06-26 11:50:15 -04002894EXPORT_SYMBOL(kern_path_create);
2895
2896struct dentry *user_path_create(int dfd, const char __user *pathname, struct path *path, int is_dir)
2897{
2898 char *tmp = getname(pathname);
2899 struct dentry *res;
2900 if (IS_ERR(tmp))
2901 return ERR_CAST(tmp);
2902 res = kern_path_create(dfd, tmp, path, is_dir);
2903 putname(tmp);
2904 return res;
2905}
2906EXPORT_SYMBOL(user_path_create);
2907
Al Viro1a67aaf2011-07-26 01:52:52 -04002908int vfs_mknod(struct inode *dir, struct dentry *dentry, umode_t mode, dev_t dev)
Linus Torvalds1da177e2005-04-16 15:20:36 -07002909{
Miklos Szeredia95164d2008-07-30 15:08:48 +02002910 int error = may_create(dir, dentry);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002911
2912 if (error)
2913 return error;
2914
Eric W. Biederman975d6b32011-11-13 12:16:43 -08002915 if ((S_ISCHR(mode) || S_ISBLK(mode)) && !capable(CAP_MKNOD))
Linus Torvalds1da177e2005-04-16 15:20:36 -07002916 return -EPERM;
2917
Al Viroacfa4382008-12-04 10:06:33 -05002918 if (!dir->i_op->mknod)
Linus Torvalds1da177e2005-04-16 15:20:36 -07002919 return -EPERM;
2920
Serge E. Hallyn08ce5f12008-04-29 01:00:10 -07002921 error = devcgroup_inode_mknod(mode, dev);
2922 if (error)
2923 return error;
2924
Linus Torvalds1da177e2005-04-16 15:20:36 -07002925 error = security_inode_mknod(dir, dentry, mode, dev);
2926 if (error)
2927 return error;
2928
Linus Torvalds1da177e2005-04-16 15:20:36 -07002929 error = dir->i_op->mknod(dir, dentry, mode, dev);
Stephen Smalleya74574a2005-09-09 13:01:44 -07002930 if (!error)
Amy Griffisf38aa942005-11-03 15:57:06 +00002931 fsnotify_create(dir, dentry);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002932 return error;
2933}
2934
Al Virof69aac02011-07-26 04:31:40 -04002935static int may_mknod(umode_t mode)
Dave Hansen463c3192008-02-15 14:37:57 -08002936{
2937 switch (mode & S_IFMT) {
2938 case S_IFREG:
2939 case S_IFCHR:
2940 case S_IFBLK:
2941 case S_IFIFO:
2942 case S_IFSOCK:
2943 case 0: /* zero mode translates to S_IFREG */
2944 return 0;
2945 case S_IFDIR:
2946 return -EPERM;
2947 default:
2948 return -EINVAL;
2949 }
2950}
2951
Al Viro8208a222011-07-25 17:32:17 -04002952SYSCALL_DEFINE4(mknodat, int, dfd, const char __user *, filename, umode_t, mode,
Heiko Carstens2e4d0922009-01-14 14:14:31 +01002953 unsigned, dev)
Linus Torvalds1da177e2005-04-16 15:20:36 -07002954{
Al Viro2ad94ae2008-07-21 09:32:51 -04002955 struct dentry *dentry;
Al Virodae6ad82011-06-26 11:50:15 -04002956 struct path path;
2957 int error;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002958
2959 if (S_ISDIR(mode))
2960 return -EPERM;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002961
Al Virodae6ad82011-06-26 11:50:15 -04002962 dentry = user_path_create(dfd, filename, &path, 0);
2963 if (IS_ERR(dentry))
2964 return PTR_ERR(dentry);
Al Viro2ad94ae2008-07-21 09:32:51 -04002965
Al Virodae6ad82011-06-26 11:50:15 -04002966 if (!IS_POSIXACL(path.dentry->d_inode))
Al Viroce3b0f82009-03-29 19:08:22 -04002967 mode &= ~current_umask();
Dave Hansen463c3192008-02-15 14:37:57 -08002968 error = may_mknod(mode);
2969 if (error)
2970 goto out_dput;
Al Virodae6ad82011-06-26 11:50:15 -04002971 error = mnt_want_write(path.mnt);
Dave Hansen463c3192008-02-15 14:37:57 -08002972 if (error)
2973 goto out_dput;
Al Virodae6ad82011-06-26 11:50:15 -04002974 error = security_path_mknod(&path, dentry, mode, dev);
Kentaro Takedabe6d3e52008-12-17 13:24:15 +09002975 if (error)
2976 goto out_drop_write;
Dave Hansen463c3192008-02-15 14:37:57 -08002977 switch (mode & S_IFMT) {
Linus Torvalds1da177e2005-04-16 15:20:36 -07002978 case 0: case S_IFREG:
Al Viro312b63f2012-06-10 18:09:36 -04002979 error = vfs_create(path.dentry->d_inode,dentry,mode,true);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002980 break;
2981 case S_IFCHR: case S_IFBLK:
Al Virodae6ad82011-06-26 11:50:15 -04002982 error = vfs_mknod(path.dentry->d_inode,dentry,mode,
Linus Torvalds1da177e2005-04-16 15:20:36 -07002983 new_decode_dev(dev));
2984 break;
2985 case S_IFIFO: case S_IFSOCK:
Al Virodae6ad82011-06-26 11:50:15 -04002986 error = vfs_mknod(path.dentry->d_inode,dentry,mode,0);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002987 break;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002988 }
Kentaro Takedabe6d3e52008-12-17 13:24:15 +09002989out_drop_write:
Al Virodae6ad82011-06-26 11:50:15 -04002990 mnt_drop_write(path.mnt);
Dave Hansen463c3192008-02-15 14:37:57 -08002991out_dput:
2992 dput(dentry);
Al Virodae6ad82011-06-26 11:50:15 -04002993 mutex_unlock(&path.dentry->d_inode->i_mutex);
2994 path_put(&path);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002995
2996 return error;
2997}
2998
Al Viro8208a222011-07-25 17:32:17 -04002999SYSCALL_DEFINE3(mknod, const char __user *, filename, umode_t, mode, unsigned, dev)
Ulrich Drepper5590ff02006-01-18 17:43:53 -08003000{
3001 return sys_mknodat(AT_FDCWD, filename, mode, dev);
3002}
3003
Al Viro18bb1db2011-07-26 01:41:39 -04003004int vfs_mkdir(struct inode *dir, struct dentry *dentry, umode_t mode)
Linus Torvalds1da177e2005-04-16 15:20:36 -07003005{
Miklos Szeredia95164d2008-07-30 15:08:48 +02003006 int error = may_create(dir, dentry);
Al Viro8de52772012-02-06 12:45:27 -05003007 unsigned max_links = dir->i_sb->s_max_links;
Linus Torvalds1da177e2005-04-16 15:20:36 -07003008
3009 if (error)
3010 return error;
3011
Al Viroacfa4382008-12-04 10:06:33 -05003012 if (!dir->i_op->mkdir)
Linus Torvalds1da177e2005-04-16 15:20:36 -07003013 return -EPERM;
3014
3015 mode &= (S_IRWXUGO|S_ISVTX);
3016 error = security_inode_mkdir(dir, dentry, mode);
3017 if (error)
3018 return error;
3019
Al Viro8de52772012-02-06 12:45:27 -05003020 if (max_links && dir->i_nlink >= max_links)
3021 return -EMLINK;
3022
Linus Torvalds1da177e2005-04-16 15:20:36 -07003023 error = dir->i_op->mkdir(dir, dentry, mode);
Stephen Smalleya74574a2005-09-09 13:01:44 -07003024 if (!error)
Amy Griffisf38aa942005-11-03 15:57:06 +00003025 fsnotify_mkdir(dir, dentry);
Linus Torvalds1da177e2005-04-16 15:20:36 -07003026 return error;
3027}
3028
Al Viroa218d0f2011-11-21 14:59:34 -05003029SYSCALL_DEFINE3(mkdirat, int, dfd, const char __user *, pathname, umode_t, mode)
Linus Torvalds1da177e2005-04-16 15:20:36 -07003030{
Dave Hansen6902d922006-09-30 23:29:01 -07003031 struct dentry *dentry;
Al Virodae6ad82011-06-26 11:50:15 -04003032 struct path path;
3033 int error;
Linus Torvalds1da177e2005-04-16 15:20:36 -07003034
Al Virodae6ad82011-06-26 11:50:15 -04003035 dentry = user_path_create(dfd, pathname, &path, 1);
Dave Hansen6902d922006-09-30 23:29:01 -07003036 if (IS_ERR(dentry))
Al Virodae6ad82011-06-26 11:50:15 -04003037 return PTR_ERR(dentry);
Dave Hansen6902d922006-09-30 23:29:01 -07003038
Al Virodae6ad82011-06-26 11:50:15 -04003039 if (!IS_POSIXACL(path.dentry->d_inode))
Al Viroce3b0f82009-03-29 19:08:22 -04003040 mode &= ~current_umask();
Al Virodae6ad82011-06-26 11:50:15 -04003041 error = mnt_want_write(path.mnt);
Dave Hansen463c3192008-02-15 14:37:57 -08003042 if (error)
3043 goto out_dput;
Al Virodae6ad82011-06-26 11:50:15 -04003044 error = security_path_mkdir(&path, dentry, mode);
Kentaro Takedabe6d3e52008-12-17 13:24:15 +09003045 if (error)
3046 goto out_drop_write;
Al Virodae6ad82011-06-26 11:50:15 -04003047 error = vfs_mkdir(path.dentry->d_inode, dentry, mode);
Kentaro Takedabe6d3e52008-12-17 13:24:15 +09003048out_drop_write:
Al Virodae6ad82011-06-26 11:50:15 -04003049 mnt_drop_write(path.mnt);
Dave Hansen463c3192008-02-15 14:37:57 -08003050out_dput:
Dave Hansen6902d922006-09-30 23:29:01 -07003051 dput(dentry);
Al Virodae6ad82011-06-26 11:50:15 -04003052 mutex_unlock(&path.dentry->d_inode->i_mutex);
3053 path_put(&path);
Linus Torvalds1da177e2005-04-16 15:20:36 -07003054 return error;
3055}
3056
Al Viroa218d0f2011-11-21 14:59:34 -05003057SYSCALL_DEFINE2(mkdir, const char __user *, pathname, umode_t, mode)
Ulrich Drepper5590ff02006-01-18 17:43:53 -08003058{
3059 return sys_mkdirat(AT_FDCWD, pathname, mode);
3060}
3061
Linus Torvalds1da177e2005-04-16 15:20:36 -07003062/*
Sage Weila71905f2011-05-24 13:06:08 -07003063 * The dentry_unhash() helper will try to drop the dentry early: we
J. Bruce Fieldsc0d02592012-02-15 11:48:40 -05003064 * should have a usage count of 1 if we're the only user of this
Sage Weila71905f2011-05-24 13:06:08 -07003065 * dentry, and if that is true (possibly after pruning the dcache),
3066 * then we drop the dentry now.
Linus Torvalds1da177e2005-04-16 15:20:36 -07003067 *
3068 * A low-level filesystem can, if it choses, legally
3069 * do a
3070 *
3071 * if (!d_unhashed(dentry))
3072 * return -EBUSY;
3073 *
3074 * if it cannot handle the case of removing a directory
3075 * that is still in use by something else..
3076 */
3077void dentry_unhash(struct dentry *dentry)
3078{
Vasily Averindc168422006-12-06 20:37:07 -08003079 shrink_dcache_parent(dentry);
Linus Torvalds1da177e2005-04-16 15:20:36 -07003080 spin_lock(&dentry->d_lock);
Sage Weil64252c72011-05-24 13:06:05 -07003081 if (dentry->d_count == 1)
Linus Torvalds1da177e2005-04-16 15:20:36 -07003082 __d_drop(dentry);
3083 spin_unlock(&dentry->d_lock);
Linus Torvalds1da177e2005-04-16 15:20:36 -07003084}
3085
3086int vfs_rmdir(struct inode *dir, struct dentry *dentry)
3087{
3088 int error = may_delete(dir, dentry, 1);
3089
3090 if (error)
3091 return error;
3092
Al Viroacfa4382008-12-04 10:06:33 -05003093 if (!dir->i_op->rmdir)
Linus Torvalds1da177e2005-04-16 15:20:36 -07003094 return -EPERM;
3095
Al Viro1d2ef592011-09-14 18:55:41 +01003096 dget(dentry);
Jes Sorensen1b1dcc12006-01-09 15:59:24 -08003097 mutex_lock(&dentry->d_inode->i_mutex);
Linus Torvalds1da177e2005-04-16 15:20:36 -07003098
Sage Weil912dbc12011-05-24 13:06:11 -07003099 error = -EBUSY;
3100 if (d_mountpoint(dentry))
3101 goto out;
3102
3103 error = security_inode_rmdir(dir, dentry);
3104 if (error)
3105 goto out;
3106
Sage Weil3cebde22011-05-29 21:20:59 -07003107 shrink_dcache_parent(dentry);
Sage Weil912dbc12011-05-24 13:06:11 -07003108 error = dir->i_op->rmdir(dir, dentry);
3109 if (error)
3110 goto out;
3111
3112 dentry->d_inode->i_flags |= S_DEAD;
3113 dont_mount(dentry);
3114
3115out:
3116 mutex_unlock(&dentry->d_inode->i_mutex);
Al Viro1d2ef592011-09-14 18:55:41 +01003117 dput(dentry);
Sage Weil912dbc12011-05-24 13:06:11 -07003118 if (!error)
3119 d_delete(dentry);
Linus Torvalds1da177e2005-04-16 15:20:36 -07003120 return error;
3121}
3122
Ulrich Drepper5590ff02006-01-18 17:43:53 -08003123static long do_rmdir(int dfd, const char __user *pathname)
Linus Torvalds1da177e2005-04-16 15:20:36 -07003124{
3125 int error = 0;
3126 char * name;
3127 struct dentry *dentry;
3128 struct nameidata nd;
3129
Al Viro2ad94ae2008-07-21 09:32:51 -04003130 error = user_path_parent(dfd, pathname, &nd, &name);
Linus Torvalds1da177e2005-04-16 15:20:36 -07003131 if (error)
Al Viro2ad94ae2008-07-21 09:32:51 -04003132 return error;
Linus Torvalds1da177e2005-04-16 15:20:36 -07003133
3134 switch(nd.last_type) {
OGAWA Hirofumi0612d9f2008-10-16 07:50:29 +09003135 case LAST_DOTDOT:
3136 error = -ENOTEMPTY;
3137 goto exit1;
3138 case LAST_DOT:
3139 error = -EINVAL;
3140 goto exit1;
3141 case LAST_ROOT:
3142 error = -EBUSY;
3143 goto exit1;
Linus Torvalds1da177e2005-04-16 15:20:36 -07003144 }
OGAWA Hirofumi0612d9f2008-10-16 07:50:29 +09003145
3146 nd.flags &= ~LOOKUP_PARENT;
3147
Jan Blunck4ac91372008-02-14 19:34:32 -08003148 mutex_lock_nested(&nd.path.dentry->d_inode->i_mutex, I_MUTEX_PARENT);
Christoph Hellwig49705b72005-11-08 21:35:06 -08003149 dentry = lookup_hash(&nd);
Linus Torvalds1da177e2005-04-16 15:20:36 -07003150 error = PTR_ERR(dentry);
Dave Hansen6902d922006-09-30 23:29:01 -07003151 if (IS_ERR(dentry))
3152 goto exit2;
Theodore Ts'oe6bc45d2011-06-06 19:19:40 -04003153 if (!dentry->d_inode) {
3154 error = -ENOENT;
3155 goto exit3;
3156 }
Dave Hansen06227532008-02-15 14:37:34 -08003157 error = mnt_want_write(nd.path.mnt);
3158 if (error)
3159 goto exit3;
Kentaro Takedabe6d3e52008-12-17 13:24:15 +09003160 error = security_path_rmdir(&nd.path, dentry);
3161 if (error)
3162 goto exit4;
Jan Blunck4ac91372008-02-14 19:34:32 -08003163 error = vfs_rmdir(nd.path.dentry->d_inode, dentry);
Kentaro Takedabe6d3e52008-12-17 13:24:15 +09003164exit4:
Dave Hansen06227532008-02-15 14:37:34 -08003165 mnt_drop_write(nd.path.mnt);
3166exit3:
Dave Hansen6902d922006-09-30 23:29:01 -07003167 dput(dentry);
3168exit2:
Jan Blunck4ac91372008-02-14 19:34:32 -08003169 mutex_unlock(&nd.path.dentry->d_inode->i_mutex);
Linus Torvalds1da177e2005-04-16 15:20:36 -07003170exit1:
Jan Blunck1d957f92008-02-14 19:34:35 -08003171 path_put(&nd.path);
Linus Torvalds1da177e2005-04-16 15:20:36 -07003172 putname(name);
3173 return error;
3174}
3175
Heiko Carstens3cdad422009-01-14 14:14:22 +01003176SYSCALL_DEFINE1(rmdir, const char __user *, pathname)
Ulrich Drepper5590ff02006-01-18 17:43:53 -08003177{
3178 return do_rmdir(AT_FDCWD, pathname);
3179}
3180
Linus Torvalds1da177e2005-04-16 15:20:36 -07003181int vfs_unlink(struct inode *dir, struct dentry *dentry)
3182{
3183 int error = may_delete(dir, dentry, 0);
3184
3185 if (error)
3186 return error;
3187
Al Viroacfa4382008-12-04 10:06:33 -05003188 if (!dir->i_op->unlink)
Linus Torvalds1da177e2005-04-16 15:20:36 -07003189 return -EPERM;
3190
Jes Sorensen1b1dcc12006-01-09 15:59:24 -08003191 mutex_lock(&dentry->d_inode->i_mutex);
Linus Torvalds1da177e2005-04-16 15:20:36 -07003192 if (d_mountpoint(dentry))
3193 error = -EBUSY;
3194 else {
3195 error = security_inode_unlink(dir, dentry);
Al Virobec10522010-03-03 14:12:08 -05003196 if (!error) {
Linus Torvalds1da177e2005-04-16 15:20:36 -07003197 error = dir->i_op->unlink(dir, dentry);
Al Virobec10522010-03-03 14:12:08 -05003198 if (!error)
Al Virod83c49f2010-04-30 17:17:09 -04003199 dont_mount(dentry);
Al Virobec10522010-03-03 14:12:08 -05003200 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07003201 }
Jes Sorensen1b1dcc12006-01-09 15:59:24 -08003202 mutex_unlock(&dentry->d_inode->i_mutex);
Linus Torvalds1da177e2005-04-16 15:20:36 -07003203
3204 /* We don't d_delete() NFS sillyrenamed files--they still exist. */
3205 if (!error && !(dentry->d_flags & DCACHE_NFSFS_RENAMED)) {
Jan Karaece95912008-02-06 01:37:13 -08003206 fsnotify_link_count(dentry->d_inode);
Linus Torvalds1da177e2005-04-16 15:20:36 -07003207 d_delete(dentry);
Linus Torvalds1da177e2005-04-16 15:20:36 -07003208 }
Robert Love0eeca282005-07-12 17:06:03 -04003209
Linus Torvalds1da177e2005-04-16 15:20:36 -07003210 return error;
3211}
3212
3213/*
3214 * Make sure that the actual truncation of the file will occur outside its
Jes Sorensen1b1dcc12006-01-09 15:59:24 -08003215 * directory's i_mutex. Truncate can take a long time if there is a lot of
Linus Torvalds1da177e2005-04-16 15:20:36 -07003216 * writeout happening, and we don't want to prevent access to the directory
3217 * while waiting on the I/O.
3218 */
Ulrich Drepper5590ff02006-01-18 17:43:53 -08003219static long do_unlinkat(int dfd, const char __user *pathname)
Linus Torvalds1da177e2005-04-16 15:20:36 -07003220{
Al Viro2ad94ae2008-07-21 09:32:51 -04003221 int error;
3222 char *name;
Linus Torvalds1da177e2005-04-16 15:20:36 -07003223 struct dentry *dentry;
3224 struct nameidata nd;
3225 struct inode *inode = NULL;
3226
Al Viro2ad94ae2008-07-21 09:32:51 -04003227 error = user_path_parent(dfd, pathname, &nd, &name);
Linus Torvalds1da177e2005-04-16 15:20:36 -07003228 if (error)
Al Viro2ad94ae2008-07-21 09:32:51 -04003229 return error;
3230
Linus Torvalds1da177e2005-04-16 15:20:36 -07003231 error = -EISDIR;
3232 if (nd.last_type != LAST_NORM)
3233 goto exit1;
OGAWA Hirofumi0612d9f2008-10-16 07:50:29 +09003234
3235 nd.flags &= ~LOOKUP_PARENT;
3236
Jan Blunck4ac91372008-02-14 19:34:32 -08003237 mutex_lock_nested(&nd.path.dentry->d_inode->i_mutex, I_MUTEX_PARENT);
Christoph Hellwig49705b72005-11-08 21:35:06 -08003238 dentry = lookup_hash(&nd);
Linus Torvalds1da177e2005-04-16 15:20:36 -07003239 error = PTR_ERR(dentry);
3240 if (!IS_ERR(dentry)) {
3241 /* Why not before? Because we want correct error value */
Török Edwin50338b82011-06-16 00:06:14 +03003242 if (nd.last.name[nd.last.len])
3243 goto slashes;
Linus Torvalds1da177e2005-04-16 15:20:36 -07003244 inode = dentry->d_inode;
Török Edwin50338b82011-06-16 00:06:14 +03003245 if (!inode)
Theodore Ts'oe6bc45d2011-06-06 19:19:40 -04003246 goto slashes;
3247 ihold(inode);
Dave Hansen06227532008-02-15 14:37:34 -08003248 error = mnt_want_write(nd.path.mnt);
3249 if (error)
3250 goto exit2;
Kentaro Takedabe6d3e52008-12-17 13:24:15 +09003251 error = security_path_unlink(&nd.path, dentry);
3252 if (error)
3253 goto exit3;
Jan Blunck4ac91372008-02-14 19:34:32 -08003254 error = vfs_unlink(nd.path.dentry->d_inode, dentry);
Kentaro Takedabe6d3e52008-12-17 13:24:15 +09003255exit3:
Dave Hansen06227532008-02-15 14:37:34 -08003256 mnt_drop_write(nd.path.mnt);
Linus Torvalds1da177e2005-04-16 15:20:36 -07003257 exit2:
3258 dput(dentry);
3259 }
Jan Blunck4ac91372008-02-14 19:34:32 -08003260 mutex_unlock(&nd.path.dentry->d_inode->i_mutex);
Linus Torvalds1da177e2005-04-16 15:20:36 -07003261 if (inode)
3262 iput(inode); /* truncate the inode here */
3263exit1:
Jan Blunck1d957f92008-02-14 19:34:35 -08003264 path_put(&nd.path);
Linus Torvalds1da177e2005-04-16 15:20:36 -07003265 putname(name);
3266 return error;
3267
3268slashes:
3269 error = !dentry->d_inode ? -ENOENT :
3270 S_ISDIR(dentry->d_inode->i_mode) ? -EISDIR : -ENOTDIR;
3271 goto exit2;
3272}
3273
Heiko Carstens2e4d0922009-01-14 14:14:31 +01003274SYSCALL_DEFINE3(unlinkat, int, dfd, const char __user *, pathname, int, flag)
Ulrich Drepper5590ff02006-01-18 17:43:53 -08003275{
3276 if ((flag & ~AT_REMOVEDIR) != 0)
3277 return -EINVAL;
3278
3279 if (flag & AT_REMOVEDIR)
3280 return do_rmdir(dfd, pathname);
3281
3282 return do_unlinkat(dfd, pathname);
3283}
3284
Heiko Carstens3480b252009-01-14 14:14:16 +01003285SYSCALL_DEFINE1(unlink, const char __user *, pathname)
Ulrich Drepper5590ff02006-01-18 17:43:53 -08003286{
3287 return do_unlinkat(AT_FDCWD, pathname);
3288}
3289
Miklos Szeredidb2e7472008-06-24 16:50:16 +02003290int vfs_symlink(struct inode *dir, struct dentry *dentry, const char *oldname)
Linus Torvalds1da177e2005-04-16 15:20:36 -07003291{
Miklos Szeredia95164d2008-07-30 15:08:48 +02003292 int error = may_create(dir, dentry);
Linus Torvalds1da177e2005-04-16 15:20:36 -07003293
3294 if (error)
3295 return error;
3296
Al Viroacfa4382008-12-04 10:06:33 -05003297 if (!dir->i_op->symlink)
Linus Torvalds1da177e2005-04-16 15:20:36 -07003298 return -EPERM;
3299
3300 error = security_inode_symlink(dir, dentry, oldname);
3301 if (error)
3302 return error;
3303
Linus Torvalds1da177e2005-04-16 15:20:36 -07003304 error = dir->i_op->symlink(dir, dentry, oldname);
Stephen Smalleya74574a2005-09-09 13:01:44 -07003305 if (!error)
Amy Griffisf38aa942005-11-03 15:57:06 +00003306 fsnotify_create(dir, dentry);
Linus Torvalds1da177e2005-04-16 15:20:36 -07003307 return error;
3308}
3309
Heiko Carstens2e4d0922009-01-14 14:14:31 +01003310SYSCALL_DEFINE3(symlinkat, const char __user *, oldname,
3311 int, newdfd, const char __user *, newname)
Linus Torvalds1da177e2005-04-16 15:20:36 -07003312{
Al Viro2ad94ae2008-07-21 09:32:51 -04003313 int error;
3314 char *from;
Dave Hansen6902d922006-09-30 23:29:01 -07003315 struct dentry *dentry;
Al Virodae6ad82011-06-26 11:50:15 -04003316 struct path path;
Linus Torvalds1da177e2005-04-16 15:20:36 -07003317
3318 from = getname(oldname);
Al Viro2ad94ae2008-07-21 09:32:51 -04003319 if (IS_ERR(from))
Linus Torvalds1da177e2005-04-16 15:20:36 -07003320 return PTR_ERR(from);
Al Viro2ad94ae2008-07-21 09:32:51 -04003321
Al Virodae6ad82011-06-26 11:50:15 -04003322 dentry = user_path_create(newdfd, newname, &path, 0);
Dave Hansen6902d922006-09-30 23:29:01 -07003323 error = PTR_ERR(dentry);
3324 if (IS_ERR(dentry))
Al Virodae6ad82011-06-26 11:50:15 -04003325 goto out_putname;
Dave Hansen6902d922006-09-30 23:29:01 -07003326
Al Virodae6ad82011-06-26 11:50:15 -04003327 error = mnt_want_write(path.mnt);
Dave Hansen75c3f292008-02-15 14:37:45 -08003328 if (error)
3329 goto out_dput;
Al Virodae6ad82011-06-26 11:50:15 -04003330 error = security_path_symlink(&path, dentry, from);
Kentaro Takedabe6d3e52008-12-17 13:24:15 +09003331 if (error)
3332 goto out_drop_write;
Al Virodae6ad82011-06-26 11:50:15 -04003333 error = vfs_symlink(path.dentry->d_inode, dentry, from);
Kentaro Takedabe6d3e52008-12-17 13:24:15 +09003334out_drop_write:
Al Virodae6ad82011-06-26 11:50:15 -04003335 mnt_drop_write(path.mnt);
Dave Hansen75c3f292008-02-15 14:37:45 -08003336out_dput:
Dave Hansen6902d922006-09-30 23:29:01 -07003337 dput(dentry);
Al Virodae6ad82011-06-26 11:50:15 -04003338 mutex_unlock(&path.dentry->d_inode->i_mutex);
3339 path_put(&path);
Dave Hansen6902d922006-09-30 23:29:01 -07003340out_putname:
Linus Torvalds1da177e2005-04-16 15:20:36 -07003341 putname(from);
3342 return error;
3343}
3344
Heiko Carstens3480b252009-01-14 14:14:16 +01003345SYSCALL_DEFINE2(symlink, const char __user *, oldname, const char __user *, newname)
Ulrich Drepper5590ff02006-01-18 17:43:53 -08003346{
3347 return sys_symlinkat(oldname, AT_FDCWD, newname);
3348}
3349
Linus Torvalds1da177e2005-04-16 15:20:36 -07003350int vfs_link(struct dentry *old_dentry, struct inode *dir, struct dentry *new_dentry)
3351{
3352 struct inode *inode = old_dentry->d_inode;
Al Viro8de52772012-02-06 12:45:27 -05003353 unsigned max_links = dir->i_sb->s_max_links;
Linus Torvalds1da177e2005-04-16 15:20:36 -07003354 int error;
3355
3356 if (!inode)
3357 return -ENOENT;
3358
Miklos Szeredia95164d2008-07-30 15:08:48 +02003359 error = may_create(dir, new_dentry);
Linus Torvalds1da177e2005-04-16 15:20:36 -07003360 if (error)
3361 return error;
3362
3363 if (dir->i_sb != inode->i_sb)
3364 return -EXDEV;
3365
3366 /*
3367 * A link to an append-only or immutable file cannot be created.
3368 */
3369 if (IS_APPEND(inode) || IS_IMMUTABLE(inode))
3370 return -EPERM;
Al Viroacfa4382008-12-04 10:06:33 -05003371 if (!dir->i_op->link)
Linus Torvalds1da177e2005-04-16 15:20:36 -07003372 return -EPERM;
Tetsuo Handa7e79eed2008-06-24 16:50:15 +02003373 if (S_ISDIR(inode->i_mode))
Linus Torvalds1da177e2005-04-16 15:20:36 -07003374 return -EPERM;
3375
3376 error = security_inode_link(old_dentry, dir, new_dentry);
3377 if (error)
3378 return error;
3379
Tetsuo Handa7e79eed2008-06-24 16:50:15 +02003380 mutex_lock(&inode->i_mutex);
Aneesh Kumar K.Vaae8a972011-01-29 18:43:27 +05303381 /* Make sure we don't allow creating hardlink to an unlinked file */
3382 if (inode->i_nlink == 0)
3383 error = -ENOENT;
Al Viro8de52772012-02-06 12:45:27 -05003384 else if (max_links && inode->i_nlink >= max_links)
3385 error = -EMLINK;
Aneesh Kumar K.Vaae8a972011-01-29 18:43:27 +05303386 else
3387 error = dir->i_op->link(old_dentry, dir, new_dentry);
Tetsuo Handa7e79eed2008-06-24 16:50:15 +02003388 mutex_unlock(&inode->i_mutex);
Stephen Smalleye31e14e2005-09-09 13:01:45 -07003389 if (!error)
Tetsuo Handa7e79eed2008-06-24 16:50:15 +02003390 fsnotify_link(dir, inode, new_dentry);
Linus Torvalds1da177e2005-04-16 15:20:36 -07003391 return error;
3392}
3393
3394/*
3395 * Hardlinks are often used in delicate situations. We avoid
3396 * security-related surprises by not following symlinks on the
3397 * newname. --KAB
3398 *
3399 * We don't follow them on the oldname either to be compatible
3400 * with linux 2.0, and to avoid hard-linking to directories
3401 * and other special files. --ADM
3402 */
Heiko Carstens2e4d0922009-01-14 14:14:31 +01003403SYSCALL_DEFINE5(linkat, int, olddfd, const char __user *, oldname,
3404 int, newdfd, const char __user *, newname, int, flags)
Linus Torvalds1da177e2005-04-16 15:20:36 -07003405{
3406 struct dentry *new_dentry;
Al Virodae6ad82011-06-26 11:50:15 -04003407 struct path old_path, new_path;
Aneesh Kumar K.V11a7b372011-01-29 18:43:42 +05303408 int how = 0;
Linus Torvalds1da177e2005-04-16 15:20:36 -07003409 int error;
Linus Torvalds1da177e2005-04-16 15:20:36 -07003410
Aneesh Kumar K.V11a7b372011-01-29 18:43:42 +05303411 if ((flags & ~(AT_SYMLINK_FOLLOW | AT_EMPTY_PATH)) != 0)
Ulrich Drepperc04030e2006-02-24 13:04:21 -08003412 return -EINVAL;
Aneesh Kumar K.V11a7b372011-01-29 18:43:42 +05303413 /*
3414 * To use null names we require CAP_DAC_READ_SEARCH
3415 * This ensures that not everyone will be able to create
3416 * handlink using the passed filedescriptor.
3417 */
3418 if (flags & AT_EMPTY_PATH) {
3419 if (!capable(CAP_DAC_READ_SEARCH))
3420 return -ENOENT;
3421 how = LOOKUP_EMPTY;
3422 }
Ulrich Drepperc04030e2006-02-24 13:04:21 -08003423
Aneesh Kumar K.V11a7b372011-01-29 18:43:42 +05303424 if (flags & AT_SYMLINK_FOLLOW)
3425 how |= LOOKUP_FOLLOW;
3426
3427 error = user_path_at(olddfd, oldname, how, &old_path);
Linus Torvalds1da177e2005-04-16 15:20:36 -07003428 if (error)
Al Viro2ad94ae2008-07-21 09:32:51 -04003429 return error;
3430
Al Virodae6ad82011-06-26 11:50:15 -04003431 new_dentry = user_path_create(newdfd, newname, &new_path, 0);
Linus Torvalds1da177e2005-04-16 15:20:36 -07003432 error = PTR_ERR(new_dentry);
Dave Hansen6902d922006-09-30 23:29:01 -07003433 if (IS_ERR(new_dentry))
Al Virodae6ad82011-06-26 11:50:15 -04003434 goto out;
3435
3436 error = -EXDEV;
3437 if (old_path.mnt != new_path.mnt)
3438 goto out_dput;
3439 error = mnt_want_write(new_path.mnt);
Dave Hansen75c3f292008-02-15 14:37:45 -08003440 if (error)
3441 goto out_dput;
Al Virodae6ad82011-06-26 11:50:15 -04003442 error = security_path_link(old_path.dentry, &new_path, new_dentry);
Kentaro Takedabe6d3e52008-12-17 13:24:15 +09003443 if (error)
3444 goto out_drop_write;
Al Virodae6ad82011-06-26 11:50:15 -04003445 error = vfs_link(old_path.dentry, new_path.dentry->d_inode, new_dentry);
Kentaro Takedabe6d3e52008-12-17 13:24:15 +09003446out_drop_write:
Al Virodae6ad82011-06-26 11:50:15 -04003447 mnt_drop_write(new_path.mnt);
Dave Hansen75c3f292008-02-15 14:37:45 -08003448out_dput:
Dave Hansen6902d922006-09-30 23:29:01 -07003449 dput(new_dentry);
Al Virodae6ad82011-06-26 11:50:15 -04003450 mutex_unlock(&new_path.dentry->d_inode->i_mutex);
3451 path_put(&new_path);
Linus Torvalds1da177e2005-04-16 15:20:36 -07003452out:
Al Viro2d8f3032008-07-22 09:59:21 -04003453 path_put(&old_path);
Linus Torvalds1da177e2005-04-16 15:20:36 -07003454
3455 return error;
3456}
3457
Heiko Carstens3480b252009-01-14 14:14:16 +01003458SYSCALL_DEFINE2(link, const char __user *, oldname, const char __user *, newname)
Ulrich Drepper5590ff02006-01-18 17:43:53 -08003459{
Ulrich Drepperc04030e2006-02-24 13:04:21 -08003460 return sys_linkat(AT_FDCWD, oldname, AT_FDCWD, newname, 0);
Ulrich Drepper5590ff02006-01-18 17:43:53 -08003461}
3462
Linus Torvalds1da177e2005-04-16 15:20:36 -07003463/*
3464 * The worst of all namespace operations - renaming directory. "Perverted"
3465 * doesn't even start to describe it. Somebody in UCB had a heck of a trip...
3466 * Problems:
3467 * a) we can get into loop creation. Check is done in is_subdir().
3468 * b) race potential - two innocent renames can create a loop together.
3469 * That's where 4.4 screws up. Current fix: serialization on
Arjan van de Vena11f3a02006-03-23 03:00:33 -08003470 * sb->s_vfs_rename_mutex. We might be more accurate, but that's another
Linus Torvalds1da177e2005-04-16 15:20:36 -07003471 * story.
3472 * c) we have to lock _three_ objects - parents and victim (if it exists).
Jes Sorensen1b1dcc12006-01-09 15:59:24 -08003473 * And that - after we got ->i_mutex on parents (until then we don't know
Linus Torvalds1da177e2005-04-16 15:20:36 -07003474 * whether the target exists). Solution: try to be smart with locking
3475 * order for inodes. We rely on the fact that tree topology may change
Arjan van de Vena11f3a02006-03-23 03:00:33 -08003476 * only under ->s_vfs_rename_mutex _and_ that parent of the object we
Linus Torvalds1da177e2005-04-16 15:20:36 -07003477 * move will be locked. Thus we can rank directories by the tree
3478 * (ancestors first) and rank all non-directories after them.
3479 * That works since everybody except rename does "lock parent, lookup,
Arjan van de Vena11f3a02006-03-23 03:00:33 -08003480 * lock child" and rename is under ->s_vfs_rename_mutex.
Linus Torvalds1da177e2005-04-16 15:20:36 -07003481 * HOWEVER, it relies on the assumption that any object with ->lookup()
3482 * has no more than 1 dentry. If "hybrid" objects will ever appear,
3483 * we'd better make sure that there's no link(2) for them.
Sage Weile4eaac02011-05-24 13:06:07 -07003484 * d) conversion from fhandle to dentry may come in the wrong moment - when
Jes Sorensen1b1dcc12006-01-09 15:59:24 -08003485 * we are removing the target. Solution: we will have to grab ->i_mutex
Linus Torvalds1da177e2005-04-16 15:20:36 -07003486 * in the fhandle_to_dentry code. [FIXME - current nfsfh.c relies on
Adam Buchbinderc41b20e2009-12-11 16:35:39 -05003487 * ->i_mutex on parents, which works but leads to some truly excessive
Linus Torvalds1da177e2005-04-16 15:20:36 -07003488 * locking].
3489 */
Adrian Bunk75c96f82005-05-05 16:16:09 -07003490static int vfs_rename_dir(struct inode *old_dir, struct dentry *old_dentry,
3491 struct inode *new_dir, struct dentry *new_dentry)
Linus Torvalds1da177e2005-04-16 15:20:36 -07003492{
3493 int error = 0;
Sage Weil9055cba2011-05-24 13:06:12 -07003494 struct inode *target = new_dentry->d_inode;
Al Viro8de52772012-02-06 12:45:27 -05003495 unsigned max_links = new_dir->i_sb->s_max_links;
Linus Torvalds1da177e2005-04-16 15:20:36 -07003496
3497 /*
3498 * If we are going to change the parent - check write permissions,
3499 * we'll need to flip '..'.
3500 */
3501 if (new_dir != old_dir) {
Al Virof419a2e2008-07-22 00:07:17 -04003502 error = inode_permission(old_dentry->d_inode, MAY_WRITE);
Linus Torvalds1da177e2005-04-16 15:20:36 -07003503 if (error)
3504 return error;
3505 }
3506
3507 error = security_inode_rename(old_dir, old_dentry, new_dir, new_dentry);
3508 if (error)
3509 return error;
3510
Al Viro1d2ef592011-09-14 18:55:41 +01003511 dget(new_dentry);
Al Virod83c49f2010-04-30 17:17:09 -04003512 if (target)
Jes Sorensen1b1dcc12006-01-09 15:59:24 -08003513 mutex_lock(&target->i_mutex);
Sage Weil9055cba2011-05-24 13:06:12 -07003514
3515 error = -EBUSY;
3516 if (d_mountpoint(old_dentry) || d_mountpoint(new_dentry))
3517 goto out;
3518
Al Viro8de52772012-02-06 12:45:27 -05003519 error = -EMLINK;
3520 if (max_links && !target && new_dir != old_dir &&
3521 new_dir->i_nlink >= max_links)
3522 goto out;
3523
Sage Weil3cebde22011-05-29 21:20:59 -07003524 if (target)
3525 shrink_dcache_parent(new_dentry);
Sage Weil9055cba2011-05-24 13:06:12 -07003526 error = old_dir->i_op->rename(old_dir, old_dentry, new_dir, new_dentry);
3527 if (error)
3528 goto out;
3529
Linus Torvalds1da177e2005-04-16 15:20:36 -07003530 if (target) {
Sage Weil9055cba2011-05-24 13:06:12 -07003531 target->i_flags |= S_DEAD;
3532 dont_mount(new_dentry);
Linus Torvalds1da177e2005-04-16 15:20:36 -07003533 }
Sage Weil9055cba2011-05-24 13:06:12 -07003534out:
3535 if (target)
3536 mutex_unlock(&target->i_mutex);
Al Viro1d2ef592011-09-14 18:55:41 +01003537 dput(new_dentry);
Stephen Smalleye31e14e2005-09-09 13:01:45 -07003538 if (!error)
Mark Fasheh349457c2006-09-08 14:22:21 -07003539 if (!(old_dir->i_sb->s_type->fs_flags & FS_RENAME_DOES_D_MOVE))
3540 d_move(old_dentry,new_dentry);
Linus Torvalds1da177e2005-04-16 15:20:36 -07003541 return error;
3542}
3543
Adrian Bunk75c96f82005-05-05 16:16:09 -07003544static int vfs_rename_other(struct inode *old_dir, struct dentry *old_dentry,
3545 struct inode *new_dir, struct dentry *new_dentry)
Linus Torvalds1da177e2005-04-16 15:20:36 -07003546{
Sage Weil51892bb2011-05-24 13:06:13 -07003547 struct inode *target = new_dentry->d_inode;
Linus Torvalds1da177e2005-04-16 15:20:36 -07003548 int error;
3549
3550 error = security_inode_rename(old_dir, old_dentry, new_dir, new_dentry);
3551 if (error)
3552 return error;
3553
3554 dget(new_dentry);
Linus Torvalds1da177e2005-04-16 15:20:36 -07003555 if (target)
Jes Sorensen1b1dcc12006-01-09 15:59:24 -08003556 mutex_lock(&target->i_mutex);
Sage Weil51892bb2011-05-24 13:06:13 -07003557
3558 error = -EBUSY;
Linus Torvalds1da177e2005-04-16 15:20:36 -07003559 if (d_mountpoint(old_dentry)||d_mountpoint(new_dentry))
Sage Weil51892bb2011-05-24 13:06:13 -07003560 goto out;
3561
3562 error = old_dir->i_op->rename(old_dir, old_dentry, new_dir, new_dentry);
3563 if (error)
3564 goto out;
3565
3566 if (target)
3567 dont_mount(new_dentry);
3568 if (!(old_dir->i_sb->s_type->fs_flags & FS_RENAME_DOES_D_MOVE))
3569 d_move(old_dentry, new_dentry);
3570out:
Linus Torvalds1da177e2005-04-16 15:20:36 -07003571 if (target)
Jes Sorensen1b1dcc12006-01-09 15:59:24 -08003572 mutex_unlock(&target->i_mutex);
Linus Torvalds1da177e2005-04-16 15:20:36 -07003573 dput(new_dentry);
3574 return error;
3575}
3576
3577int vfs_rename(struct inode *old_dir, struct dentry *old_dentry,
3578 struct inode *new_dir, struct dentry *new_dentry)
3579{
3580 int error;
3581 int is_dir = S_ISDIR(old_dentry->d_inode->i_mode);
Eric Paris59b0df22010-02-08 12:53:52 -05003582 const unsigned char *old_name;
Linus Torvalds1da177e2005-04-16 15:20:36 -07003583
3584 if (old_dentry->d_inode == new_dentry->d_inode)
3585 return 0;
3586
3587 error = may_delete(old_dir, old_dentry, is_dir);
3588 if (error)
3589 return error;
3590
3591 if (!new_dentry->d_inode)
Miklos Szeredia95164d2008-07-30 15:08:48 +02003592 error = may_create(new_dir, new_dentry);
Linus Torvalds1da177e2005-04-16 15:20:36 -07003593 else
3594 error = may_delete(new_dir, new_dentry, is_dir);
3595 if (error)
3596 return error;
3597
Al Viroacfa4382008-12-04 10:06:33 -05003598 if (!old_dir->i_op->rename)
Linus Torvalds1da177e2005-04-16 15:20:36 -07003599 return -EPERM;
3600
Robert Love0eeca282005-07-12 17:06:03 -04003601 old_name = fsnotify_oldname_init(old_dentry->d_name.name);
3602
Linus Torvalds1da177e2005-04-16 15:20:36 -07003603 if (is_dir)
3604 error = vfs_rename_dir(old_dir,old_dentry,new_dir,new_dentry);
3605 else
3606 error = vfs_rename_other(old_dir,old_dentry,new_dir,new_dentry);
Al Viro123df292009-12-25 04:57:57 -05003607 if (!error)
3608 fsnotify_move(old_dir, new_dir, old_name, is_dir,
Al Viro5a190ae2007-06-07 12:19:32 -04003609 new_dentry->d_inode, old_dentry);
Robert Love0eeca282005-07-12 17:06:03 -04003610 fsnotify_oldname_free(old_name);
3611
Linus Torvalds1da177e2005-04-16 15:20:36 -07003612 return error;
3613}
3614
Heiko Carstens2e4d0922009-01-14 14:14:31 +01003615SYSCALL_DEFINE4(renameat, int, olddfd, const char __user *, oldname,
3616 int, newdfd, const char __user *, newname)
Linus Torvalds1da177e2005-04-16 15:20:36 -07003617{
Al Viro2ad94ae2008-07-21 09:32:51 -04003618 struct dentry *old_dir, *new_dir;
3619 struct dentry *old_dentry, *new_dentry;
3620 struct dentry *trap;
Linus Torvalds1da177e2005-04-16 15:20:36 -07003621 struct nameidata oldnd, newnd;
Al Viro2ad94ae2008-07-21 09:32:51 -04003622 char *from;
3623 char *to;
3624 int error;
Linus Torvalds1da177e2005-04-16 15:20:36 -07003625
Al Viro2ad94ae2008-07-21 09:32:51 -04003626 error = user_path_parent(olddfd, oldname, &oldnd, &from);
Linus Torvalds1da177e2005-04-16 15:20:36 -07003627 if (error)
3628 goto exit;
3629
Al Viro2ad94ae2008-07-21 09:32:51 -04003630 error = user_path_parent(newdfd, newname, &newnd, &to);
Linus Torvalds1da177e2005-04-16 15:20:36 -07003631 if (error)
3632 goto exit1;
3633
3634 error = -EXDEV;
Jan Blunck4ac91372008-02-14 19:34:32 -08003635 if (oldnd.path.mnt != newnd.path.mnt)
Linus Torvalds1da177e2005-04-16 15:20:36 -07003636 goto exit2;
3637
Jan Blunck4ac91372008-02-14 19:34:32 -08003638 old_dir = oldnd.path.dentry;
Linus Torvalds1da177e2005-04-16 15:20:36 -07003639 error = -EBUSY;
3640 if (oldnd.last_type != LAST_NORM)
3641 goto exit2;
3642
Jan Blunck4ac91372008-02-14 19:34:32 -08003643 new_dir = newnd.path.dentry;
Linus Torvalds1da177e2005-04-16 15:20:36 -07003644 if (newnd.last_type != LAST_NORM)
3645 goto exit2;
3646
OGAWA Hirofumi0612d9f2008-10-16 07:50:29 +09003647 oldnd.flags &= ~LOOKUP_PARENT;
3648 newnd.flags &= ~LOOKUP_PARENT;
OGAWA Hirofumi4e9ed2f2008-10-16 07:50:29 +09003649 newnd.flags |= LOOKUP_RENAME_TARGET;
OGAWA Hirofumi0612d9f2008-10-16 07:50:29 +09003650
Linus Torvalds1da177e2005-04-16 15:20:36 -07003651 trap = lock_rename(new_dir, old_dir);
3652
Christoph Hellwig49705b72005-11-08 21:35:06 -08003653 old_dentry = lookup_hash(&oldnd);
Linus Torvalds1da177e2005-04-16 15:20:36 -07003654 error = PTR_ERR(old_dentry);
3655 if (IS_ERR(old_dentry))
3656 goto exit3;
3657 /* source must exist */
3658 error = -ENOENT;
3659 if (!old_dentry->d_inode)
3660 goto exit4;
3661 /* unless the source is a directory trailing slashes give -ENOTDIR */
3662 if (!S_ISDIR(old_dentry->d_inode->i_mode)) {
3663 error = -ENOTDIR;
3664 if (oldnd.last.name[oldnd.last.len])
3665 goto exit4;
3666 if (newnd.last.name[newnd.last.len])
3667 goto exit4;
3668 }
3669 /* source should not be ancestor of target */
3670 error = -EINVAL;
3671 if (old_dentry == trap)
3672 goto exit4;
Christoph Hellwig49705b72005-11-08 21:35:06 -08003673 new_dentry = lookup_hash(&newnd);
Linus Torvalds1da177e2005-04-16 15:20:36 -07003674 error = PTR_ERR(new_dentry);
3675 if (IS_ERR(new_dentry))
3676 goto exit4;
3677 /* target should not be an ancestor of source */
3678 error = -ENOTEMPTY;
3679 if (new_dentry == trap)
3680 goto exit5;
3681
Dave Hansen9079b1e2008-02-15 14:37:49 -08003682 error = mnt_want_write(oldnd.path.mnt);
3683 if (error)
3684 goto exit5;
Kentaro Takedabe6d3e52008-12-17 13:24:15 +09003685 error = security_path_rename(&oldnd.path, old_dentry,
3686 &newnd.path, new_dentry);
3687 if (error)
3688 goto exit6;
Linus Torvalds1da177e2005-04-16 15:20:36 -07003689 error = vfs_rename(old_dir->d_inode, old_dentry,
3690 new_dir->d_inode, new_dentry);
Kentaro Takedabe6d3e52008-12-17 13:24:15 +09003691exit6:
Dave Hansen9079b1e2008-02-15 14:37:49 -08003692 mnt_drop_write(oldnd.path.mnt);
Linus Torvalds1da177e2005-04-16 15:20:36 -07003693exit5:
3694 dput(new_dentry);
3695exit4:
3696 dput(old_dentry);
3697exit3:
3698 unlock_rename(new_dir, old_dir);
3699exit2:
Jan Blunck1d957f92008-02-14 19:34:35 -08003700 path_put(&newnd.path);
Al Viro2ad94ae2008-07-21 09:32:51 -04003701 putname(to);
Linus Torvalds1da177e2005-04-16 15:20:36 -07003702exit1:
Jan Blunck1d957f92008-02-14 19:34:35 -08003703 path_put(&oldnd.path);
Linus Torvalds1da177e2005-04-16 15:20:36 -07003704 putname(from);
Al Viro2ad94ae2008-07-21 09:32:51 -04003705exit:
Linus Torvalds1da177e2005-04-16 15:20:36 -07003706 return error;
3707}
3708
Heiko Carstensa26eab22009-01-14 14:14:17 +01003709SYSCALL_DEFINE2(rename, const char __user *, oldname, const char __user *, newname)
Ulrich Drepper5590ff02006-01-18 17:43:53 -08003710{
3711 return sys_renameat(AT_FDCWD, oldname, AT_FDCWD, newname);
3712}
3713
Linus Torvalds1da177e2005-04-16 15:20:36 -07003714int vfs_readlink(struct dentry *dentry, char __user *buffer, int buflen, const char *link)
3715{
3716 int len;
3717
3718 len = PTR_ERR(link);
3719 if (IS_ERR(link))
3720 goto out;
3721
3722 len = strlen(link);
3723 if (len > (unsigned) buflen)
3724 len = buflen;
3725 if (copy_to_user(buffer, link, len))
3726 len = -EFAULT;
3727out:
3728 return len;
3729}
3730
3731/*
3732 * A helper for ->readlink(). This should be used *ONLY* for symlinks that
3733 * have ->follow_link() touching nd only in nd_set_link(). Using (or not
3734 * using) it for any given inode is up to filesystem.
3735 */
3736int generic_readlink(struct dentry *dentry, char __user *buffer, int buflen)
3737{
3738 struct nameidata nd;
Linus Torvaldscc314ee2005-08-19 18:02:56 -07003739 void *cookie;
Marcin Slusarz694a1762008-06-09 16:40:37 -07003740 int res;
Linus Torvaldscc314ee2005-08-19 18:02:56 -07003741
Linus Torvalds1da177e2005-04-16 15:20:36 -07003742 nd.depth = 0;
Linus Torvaldscc314ee2005-08-19 18:02:56 -07003743 cookie = dentry->d_inode->i_op->follow_link(dentry, &nd);
Marcin Slusarz694a1762008-06-09 16:40:37 -07003744 if (IS_ERR(cookie))
3745 return PTR_ERR(cookie);
3746
3747 res = vfs_readlink(dentry, buffer, buflen, nd_get_link(&nd));
3748 if (dentry->d_inode->i_op->put_link)
3749 dentry->d_inode->i_op->put_link(dentry, &nd, cookie);
3750 return res;
Linus Torvalds1da177e2005-04-16 15:20:36 -07003751}
3752
3753int vfs_follow_link(struct nameidata *nd, const char *link)
3754{
3755 return __vfs_follow_link(nd, link);
3756}
3757
3758/* get the link contents into pagecache */
3759static char *page_getlink(struct dentry * dentry, struct page **ppage)
3760{
Duane Griffinebd09ab2008-12-19 20:47:12 +00003761 char *kaddr;
3762 struct page *page;
Linus Torvalds1da177e2005-04-16 15:20:36 -07003763 struct address_space *mapping = dentry->d_inode->i_mapping;
Pekka Enberg090d2b12006-06-23 02:05:08 -07003764 page = read_mapping_page(mapping, 0, NULL);
Linus Torvalds1da177e2005-04-16 15:20:36 -07003765 if (IS_ERR(page))
Nick Piggin6fe69002007-05-06 14:49:04 -07003766 return (char*)page;
Linus Torvalds1da177e2005-04-16 15:20:36 -07003767 *ppage = page;
Duane Griffinebd09ab2008-12-19 20:47:12 +00003768 kaddr = kmap(page);
3769 nd_terminate_link(kaddr, dentry->d_inode->i_size, PAGE_SIZE - 1);
3770 return kaddr;
Linus Torvalds1da177e2005-04-16 15:20:36 -07003771}
3772
3773int page_readlink(struct dentry *dentry, char __user *buffer, int buflen)
3774{
3775 struct page *page = NULL;
3776 char *s = page_getlink(dentry, &page);
3777 int res = vfs_readlink(dentry,buffer,buflen,s);
3778 if (page) {
3779 kunmap(page);
3780 page_cache_release(page);
3781 }
3782 return res;
3783}
3784
Linus Torvaldscc314ee2005-08-19 18:02:56 -07003785void *page_follow_link_light(struct dentry *dentry, struct nameidata *nd)
Linus Torvalds1da177e2005-04-16 15:20:36 -07003786{
Linus Torvaldscc314ee2005-08-19 18:02:56 -07003787 struct page *page = NULL;
Linus Torvalds1da177e2005-04-16 15:20:36 -07003788 nd_set_link(nd, page_getlink(dentry, &page));
Linus Torvaldscc314ee2005-08-19 18:02:56 -07003789 return page;
Linus Torvalds1da177e2005-04-16 15:20:36 -07003790}
3791
Linus Torvaldscc314ee2005-08-19 18:02:56 -07003792void page_put_link(struct dentry *dentry, struct nameidata *nd, void *cookie)
Linus Torvalds1da177e2005-04-16 15:20:36 -07003793{
Linus Torvaldscc314ee2005-08-19 18:02:56 -07003794 struct page *page = cookie;
3795
3796 if (page) {
Linus Torvalds1da177e2005-04-16 15:20:36 -07003797 kunmap(page);
3798 page_cache_release(page);
Linus Torvalds1da177e2005-04-16 15:20:36 -07003799 }
3800}
3801
Nick Piggin54566b22009-01-04 12:00:53 -08003802/*
3803 * The nofs argument instructs pagecache_write_begin to pass AOP_FLAG_NOFS
3804 */
3805int __page_symlink(struct inode *inode, const char *symname, int len, int nofs)
Linus Torvalds1da177e2005-04-16 15:20:36 -07003806{
3807 struct address_space *mapping = inode->i_mapping;
Kirill Korotaev0adb25d2006-03-11 03:27:13 -08003808 struct page *page;
Nick Pigginafddba42007-10-16 01:25:01 -07003809 void *fsdata;
Dmitriy Monakhovbeb497a2007-02-16 01:27:18 -08003810 int err;
Linus Torvalds1da177e2005-04-16 15:20:36 -07003811 char *kaddr;
Nick Piggin54566b22009-01-04 12:00:53 -08003812 unsigned int flags = AOP_FLAG_UNINTERRUPTIBLE;
3813 if (nofs)
3814 flags |= AOP_FLAG_NOFS;
Linus Torvalds1da177e2005-04-16 15:20:36 -07003815
NeilBrown7e53cac2006-03-25 03:07:57 -08003816retry:
Nick Pigginafddba42007-10-16 01:25:01 -07003817 err = pagecache_write_begin(NULL, mapping, 0, len-1,
Nick Piggin54566b22009-01-04 12:00:53 -08003818 flags, &page, &fsdata);
Linus Torvalds1da177e2005-04-16 15:20:36 -07003819 if (err)
Nick Pigginafddba42007-10-16 01:25:01 -07003820 goto fail;
3821
Cong Wange8e3c3d2011-11-25 23:14:27 +08003822 kaddr = kmap_atomic(page);
Linus Torvalds1da177e2005-04-16 15:20:36 -07003823 memcpy(kaddr, symname, len-1);
Cong Wange8e3c3d2011-11-25 23:14:27 +08003824 kunmap_atomic(kaddr);
Nick Pigginafddba42007-10-16 01:25:01 -07003825
3826 err = pagecache_write_end(NULL, mapping, 0, len-1, len-1,
3827 page, fsdata);
Linus Torvalds1da177e2005-04-16 15:20:36 -07003828 if (err < 0)
3829 goto fail;
Nick Pigginafddba42007-10-16 01:25:01 -07003830 if (err < len-1)
3831 goto retry;
3832
Linus Torvalds1da177e2005-04-16 15:20:36 -07003833 mark_inode_dirty(inode);
3834 return 0;
Linus Torvalds1da177e2005-04-16 15:20:36 -07003835fail:
3836 return err;
3837}
3838
Kirill Korotaev0adb25d2006-03-11 03:27:13 -08003839int page_symlink(struct inode *inode, const char *symname, int len)
3840{
3841 return __page_symlink(inode, symname, len,
Nick Piggin54566b22009-01-04 12:00:53 -08003842 !(mapping_gfp_mask(inode->i_mapping) & __GFP_FS));
Kirill Korotaev0adb25d2006-03-11 03:27:13 -08003843}
3844
Arjan van de Ven92e1d5b2007-02-12 00:55:39 -08003845const struct inode_operations page_symlink_inode_operations = {
Linus Torvalds1da177e2005-04-16 15:20:36 -07003846 .readlink = generic_readlink,
3847 .follow_link = page_follow_link_light,
3848 .put_link = page_put_link,
3849};
3850
Al Viro2d8f3032008-07-22 09:59:21 -04003851EXPORT_SYMBOL(user_path_at);
David Howellscc53ce52011-01-14 18:45:26 +00003852EXPORT_SYMBOL(follow_down_one);
Linus Torvalds1da177e2005-04-16 15:20:36 -07003853EXPORT_SYMBOL(follow_down);
3854EXPORT_SYMBOL(follow_up);
3855EXPORT_SYMBOL(get_write_access); /* binfmt_aout */
3856EXPORT_SYMBOL(getname);
3857EXPORT_SYMBOL(lock_rename);
Linus Torvalds1da177e2005-04-16 15:20:36 -07003858EXPORT_SYMBOL(lookup_one_len);
3859EXPORT_SYMBOL(page_follow_link_light);
3860EXPORT_SYMBOL(page_put_link);
3861EXPORT_SYMBOL(page_readlink);
Kirill Korotaev0adb25d2006-03-11 03:27:13 -08003862EXPORT_SYMBOL(__page_symlink);
Linus Torvalds1da177e2005-04-16 15:20:36 -07003863EXPORT_SYMBOL(page_symlink);
3864EXPORT_SYMBOL(page_symlink_inode_operations);
Al Virod1811462008-08-02 00:49:18 -04003865EXPORT_SYMBOL(kern_path);
Josef 'Jeff' Sipek16f18202007-07-19 01:48:18 -07003866EXPORT_SYMBOL(vfs_path_lookup);
Al Virof419a2e2008-07-22 00:07:17 -04003867EXPORT_SYMBOL(inode_permission);
Linus Torvalds1da177e2005-04-16 15:20:36 -07003868EXPORT_SYMBOL(unlock_rename);
3869EXPORT_SYMBOL(vfs_create);
3870EXPORT_SYMBOL(vfs_follow_link);
3871EXPORT_SYMBOL(vfs_link);
3872EXPORT_SYMBOL(vfs_mkdir);
3873EXPORT_SYMBOL(vfs_mknod);
3874EXPORT_SYMBOL(generic_permission);
3875EXPORT_SYMBOL(vfs_readlink);
3876EXPORT_SYMBOL(vfs_rename);
3877EXPORT_SYMBOL(vfs_rmdir);
3878EXPORT_SYMBOL(vfs_symlink);
3879EXPORT_SYMBOL(vfs_unlink);
3880EXPORT_SYMBOL(dentry_unhash);
3881EXPORT_SYMBOL(generic_readlink);